Zen Image Edit: Qwen-Image-2.1 on Qwen3.5-0.8B with a text-fusion adapter inside the DiT

Text encoder replaced (Qwen3-VL-8B -> Qwen3.5-0.8B + 158M adapter), adapter embedded in the
transformer as text_fusion before txt_in. Weights fp16 (VAE fp32, 13 files via LFS), 8 example
images generated by this pipeline, NOTICE/LICENSE included per the Qwen Research License.

.gitattributes

@@ -33,3 +33,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text
+*.jpg filter=lfs diff=lfs merge=lfs -text
+*.jpeg filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,3 @@
+__pycache__/
+*.pyc
+.ipynb_checkpoints/

LICENSE

@@ -0,0 +1,55 @@
+Qwen RESEARCH LICENSE AGREEMENT
+
+Qwen RESEARCH LICENSE AGREEMENT Release Date: September 20, 2026
+
+By clicking to agree or by using or distributing any portion or element of the Qwen Materials, you will be deemed to have recognized and accepted the content of this Agreement, which is effective immediately.
+
+1. Definitions
+    a. This Qwen RESEARCH LICENSE AGREEMENT (this "Agreement") shall mean the terms and conditions for use, reproduction, distribution and modification of the Materials as defined by this Agreement.
+    b. "We" (or "Us") shall mean Hangzhou Tongyi Laboratory Technology Co., Ltd.
+    c. "You" (or "Your") shall mean a natural person or legal entity exercising the rights granted by this Agreement and/or using the Materials for any purpose and in any field of use.
+    d. "Third Parties" shall mean individuals or legal entities that are not under common control with us or you.
+    e. "Qwen" shall mean the large language models, diffusion models, and software and algorithms, consisting of trained model weights, parameters (including optimizer states), machine-learning model code, inference-enabling code, training-enabling code, fine-tuning enabling code and other elements of the foregoing distributed by us.
+    f. "Materials" shall mean, collectively, our proprietary Qwen and Documentation (and any portion thereof) made available under this Agreement.
+    g. "Source" form shall mean the preferred form for making modifications, including but not limited to model source code, documentation source, and configuration files.
+    h. "Object" form shall mean any form resulting from mechanical transformation or translation of a Source form, including but not limited to compiled object code, generated documentation, and conversions to other media types.
+    i. "Non-Commercial" shall mean for research or evaluation purposes only.
+
+2. Grant of Rights
+    a. You are granted a non-exclusive, worldwide, non-transferable and royalty-free limited license under our intellectual property or other rights owned by us embodied in the Materials to use, reproduce, distribute, copy, create derivative works of, and make modifications to the Materials FOR NON-COMMERCIAL PURPOSES ONLY. 
+    b. You shall not use the Materials for any commercial purpose without obtaining a separate commercial license from us. If you wish to use the Materials commercially, you shall request a license from us at model-business@notice.qwencloud.com.
+
+3. Redistribution
+Subject to Section 2 (Grant of Rights), you may distribute copies or make the Materials, or derivative works thereof, available as part of a product or service that contains any of them, with or without modifications, and in Source or Object form, provided that you meet the following conditions:
+    a. You shall give any other recipients of the Materials or derivative works a copy of this Agreement;
+    b. You shall cause any modified files to carry prominent notices stating that you changed the files;
+    c. You shall retain in all copies of the Materials that you distribute the following attribution notices within a "Notice" text file distributed as a part of such copies: "Qwen is licensed under the Qwen RESEARCH LICENSE AGREEMENT, Copyright (c) 2026 Hangzhou Tongyi Laboratory Technology Co., Ltd. All Rights Reserved."; and
+    d. You may add your own copyright statement to your modifications and may provide additional or different license terms and conditions for use, reproduction, or distribution of your modifications, or for any such derivative works as a whole, provided your use, reproduction, and distribution of the work otherwise complies with the terms and conditions of this Agreement.
+
+4. Rules of use
+    a. The Materials may be subject to export controls or restrictions in China, the United States or other countries or regions. You shall comply with applicable laws and regulations in your use of the Materials.
+    b. If you use the Materials or any outputs or results therefrom to create, train, fine-tune, or improve an AI model that is distributed or made available, you shall prominently display “Built with Qwen” or “Improved using Qwen” in the related product documentation.
+    c. You shall not use "Qwen" as the primary name or identifier of any derivative works or products; reasonable descriptive use (e.g., "fine-tuned from Qwen Image") is permitted.
+
+5. Intellectual Property
+    a. We retain ownership of all intellectual property rights in and to the Materials and derivatives made by or for us. Conditioned upon compliance with the terms and conditions of this Agreement, with respect to any derivative works and modifications of the Materials that are made by you, you are and will be the owner of such derivative works and modifications.
+    b. No trademark license is granted to use the trade names, trademarks, service marks, or product names of us, except as required to fulfill notice requirements under this Agreement or as required for reasonable and customary use in describing and redistributing the Materials.
+    c. If you commence a lawsuit or other proceedings (including a cross-claim or counterclaim in a lawsuit) against us or any entity alleging that the Materials or any output therefrom, or any part of the foregoing, infringe any intellectual property or other right owned or licensable by you, then all licenses granted to you under this Agreement shall terminate as of the date such lawsuit or other proceeding is commenced or brought.
+
+6. Disclaimer of Warranty and Limitation of Liability
+    a. We are not obligated to support, update, provide training for, or develop any further version of the Qwen Materials or to grant any license thereto.
+    b. THE MATERIALS ARE PROVIDED "AS IS" WITHOUT ANY EXPRESS OR IMPLIED WARRANTY OF ANY KIND INCLUDING WARRANTIES OF MERCHANTABILITY, NONINFRINGEMENT, OR FITNESS FOR A PARTICULAR PURPOSE. WE MAKE NO WARRANTY AND ASSUME NO RESPONSIBILITY FOR THE SAFETY OR STABILITY OF THE MATERIALS AND ANY OUTPUT THEREFROM.
+    c. IN NO EVENT SHALL WE BE LIABLE TO YOU FOR ANY DAMAGES, INCLUDING, BUT NOT LIMITED TO ANY DIRECT, OR INDIRECT, SPECIAL OR CONSEQUENTIAL DAMAGES ARISING FROM YOUR USE OR INABILITY TO USE THE MATERIALS OR ANY OUTPUT OF IT, NO MATTER HOW IT’S CAUSED.
+    d. You will defend, indemnify and hold harmless us from and against any claim by any third party arising out of or related to your use or distribution of the Materials.
+
+7. Survival and Termination.
+    a. The term of this Agreement shall commence upon your acceptance of this Agreement or access to the Materials and will continue in full force and effect until terminated in accordance with the terms and conditions herein.
+    b. We may terminate this Agreement if you breach any of the terms or conditions of this Agreement. Upon termination of this Agreement, you must delete and cease use of the Materials. Sections 6 and 8 shall survive the termination of this Agreement.
+
+8. Governing Law and Jurisdiction.
+    a. This Agreement and any dispute arising out of or relating to it will be governed by the laws of China, without regard to conflict of law principles, and the UN Convention on Contracts for the International Sale of Goods does not apply to this Agreement.
+    b. The People's Courts in Hangzhou City shall have exclusive jurisdiction over any dispute arising out of this Agreement.
+
+9. Other Terms and Conditions.
+    a. Any arrangements, understandings, or agreements regarding the Material not stated herein are separate from and independent of the terms and conditions of this Agreement. You shall request a separate license from us, if you use the Materials in ways not expressly agreed to in this Agreement. 
+    b. We shall not be bound by any additional or different terms or conditions communicated by you unless expressly agreed.

LICENSE-Qwen3.5-0.8B

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Alibaba Cloud
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

NOTICE

@@ -0,0 +1,26 @@
+NOTICE
+======
+
+Qwen is licensed under the Qwen RESEARCH LICENSE AGREEMENT, Copyright (c) 2026
+Hangzhou Tongyi Laboratory Technology Co., Ltd. All Rights Reserved.
+
+This repository is a derivative work of Qwen-Image-2.1. The full agreement is in
+`LICENSE`, a copy of it is given to every recipient of these files.
+
+Modified files, as required by section 2.b of the agreement:
+
+  transformer/config.json    added the `text_fusion_config` key, `_class_name` changed to
+                             QwenImage21FusionTransformer2DModel
+  transformer/*.safetensors  re-saved in fp16, plus the 66 `text_fusion.*` tensors of the
+                             adapter from the image21-08b-text-encoder-adapter project (adapter_v11)
+  vae/                       copied unchanged (fp32)
+  scheduler/                 copied unchanged
+  text_encoder/              different model: Qwen3.5-0.8B in fp16 instead of Qwen3-VL-8B
+  tokenizer/, processor/     different tokenizer/processor: the Qwen3.5-0.8B ones
+
+Files added by this repository: pipeline.py, transformer.py, example.py, model_index.json,
+media/, README.md.
+
+Improved using Qwen.
+The Qwen3.5-0.8B text encoder is redistributed under the Apache License 2.0, see
+`LICENSE-Qwen3.5-0.8B`.

README.md

@@ -0,0 +1,147 @@
+---
+license: other
+license_name: qwen-research
+license_link: https://huggingface.co/AiArtLab/zen-image-edit/blob/main/LICENSE
+library_name: diffusers
+pipeline_tag: image-to-image
+base_model:
+  - Qwen/Qwen-Image-2.1
+  - Qwen/Qwen3.5-0.8B
+tags:
+  - text-to-image
+  - image-editing
+  - diffusers
+  - qwen-image
+  - text-encoder
+  - adapter
+---
+
+# Zen Image Edit
+
+*Qwen-Image-2.1 on a 0.8B text encoder.*
+Text-to-image, character and scene editing, and transparent (RGBA) generation in one pipeline.
+
+<img src="media/hero.jpg" width="512"/>
+
+| | |
+|---|---|
+| transformer | Qwen-Image-2.1 DiT — 32 layers, 14.5 GB fp16, plus a **158M text-fusion adapter** inside |
+| text encoder | **Qwen3.5-0.8B**, 1.7 GB fp16 (native: Qwen3-VL-8B, 17.5 GB) |
+| conditioning | cosine **0.94** against the native Qwen3-VL-8B encoder (text positions) |
+| VAE | Qwen-Image-2.1, 16× spatial, fp32 |
+| scheduler | `FlowMatchEulerDiscreteScheduler` |
+| resolution | `output_resolution`, 1024 by default; follows the condition image aspect ratio |
+| precision | fp16 everywhere except the VAE |
+| peak VRAM | ~17.5 GB resident, less with `enable_model_cpu_offload()` |
+
+### What changed
+
+The text encoder is replaced by **Qwen3.5-0.8B** plus a 158M adapter, fine-tuned to reproduce what
+the native encoder produced — both from plain text and from text read together with the reference
+images (**Improved using Qwen**). The adapter lives *inside* the DiT as its text-fusion block, so the
+whole model is one self-contained diffusers folder and no 17.5 GB encoder is needed anywhere.
+
+### Examples
+
+Every image below is generated by this pipeline with 30 steps at 1024 px.
+
+**Text-to-image**
+
+| | |
+|---|---|
+| ![t2i](media/t2i.jpg) | ![hero](media/hero.jpg) |
+
+**Edit — one condition image** (background change, subject kept)
+
+![edit single](media/edit_single.jpg)
+
+**Edit — two condition images** (character replacement: identity from `<image1>`, pose/clothing/scene from `<image2>`)
+
+| | |
+|---|---|
+| ![edit swap](media/edit_swap.jpg) | ![edit char](media/edit_char.jpg) |
+
+**Edit — three condition images** (subject from `<image1>`, scene from `<image2>`, lighting from `<image3>`)
+
+![edit three](media/edit_three.jpg)
+
+**Transparent RGBA**
+
+![transparent](media/transparent.png)
+
+### Usage
+
+```python
+import torch
+from pipeline import ZenImageEditPipeline      # shipped in this repo
+
+pipe = ZenImageEditPipeline.from_pretrained(".", dtype=torch.float16)
+pipe.enable_model_cpu_offload()                # 14.5 GB DiT + fp32 VAE decoder do not co-reside on 32 GB
+
+# text-to-image
+image = pipe(prompt="a red fox in a snowy forest at dusk, cinematic, 85mm",
+             output_resolution=1024, num_inference_steps=30,
+             generator=torch.Generator("cuda").manual_seed(1234)).images[0]
+
+# editing: 1..N condition images, referenced in the prompt by TAG <image1>, <image2>, ...
+image = pipe(prompt="Replace the woman in <image2> with the woman from <image1>; keep <image2> pose, "
+                    "clothing and background unchanged.",
+             image=[ref_image, scene_image],
+             output_resolution=1024, num_inference_steps=30,
+             generator=torch.Generator("cuda").manual_seed(1234)).images[0]
+```
+
+CLI: `python example.py --prompt "..." [--image a.png b.png] --out out.png`
+
+### Files
+
+```
+pipeline.py          ZenImageEditPipeline — one class for t2i and editing, as QwenImage21Pipeline
+transformer.py       QwenImage21FusionTransformer2DModel + the text-fusion blocks
+example.py           CLI for both modes
+transformer/         DiT config + 2 fp16 shards, adapter merged in as text_fusion.*
+text_encoder/        Qwen3.5-0.8B, fp16
+processor/           its processor (image slicing + tokenization)
+tokenizer/           its tokenizer
+vae/                 Qwen-Image-2.1 VAE, fp32
+scheduler/           FlowMatchEulerDiscreteScheduler config
+media/               the examples above
+```
+
+`QwenImage21FusionTransformer2DModel` is a custom class defined in `transformer.py`, not registered
+inside `diffusers`, so plain `DiffusionPipeline.from_pretrained` does not resolve it. Load through the
+shipped pipeline with this folder on `sys.path`.
+
+### Limitations
+
+* **English only** — that is all the adapter was trained and tested on; other languages drift.
+* **Numerals on signage** come out wrong: "OPEN 24 HOURS" renders as "OPEN **26** HOURS" on every
+  seed tried. Words are fine. ![numbers](media/limit_numbers.jpg)
+* Batch size >1 at 1024 px peaks near 28 GB; one prompt per call is the safe mode.
+
+### NOTICE
+
+Qwen is licensed under the Qwen RESEARCH LICENSE AGREEMENT, Copyright (c) 2026 Hangzhou Tongyi
+Laboratory Technology Co., Ltd. All Rights Reserved.
+
+This is a derivative work of Qwen-Image-2.1 — the full agreement is in `LICENSE`, the list of
+modified files and the remainder of the required attribution is in `NOTICE`. The Qwen3.5-0.8B text
+encoder is redistributed under the Apache License 2.0, see `LICENSE-Qwen3.5-0.8B`.
+
+## Contacts
+
+Please contact with us if you may provide some GPU's or money on training
+
+ - telegram [recoilme](https://t.me/recoilme) *prefered way
+ - mail at aiartlab.org (slow response)
+
+## Citation
+
+```bibtex
+@misc{zenimageedit,
+  title={Zen Image Edit},
+  author={recoilme and AiArtLab Team},
+  url={https://huggingface.co/AiArtLab/zen-image-edit},
+  year={2026}
+}
+```

example.py

@@ -0,0 +1,68 @@
+#!/usr/bin/env python3
+"""zen-image-edit inference: no native text encoder (Qwen3-VL-8B, 17.5 GB) anywhere.
+
+On the GPU: Qwen3.5-0.8B (~1.7 GB), the DiT with the adapter inside (~14.5 GB) and the VAE
+(~1.4 GB, fp32).
+
+  # text-to-image
+  python example.py --prompt "a red fox in a snowy forest at dusk, cinematic, 85mm" --out fox.png
+
+  # editing: 1..N condition images, referenced in the prompt by TAG <image1>, <image2>, ...
+  python example.py --image ref.png scene.png \
+      --prompt "Replace the woman in <image2> with the woman from <image1>; keep <image2> pose, \\
+clothing and background unchanged." --out swap.png
+"""
+import argparse
+import os
+import sys
+
+import torch
+from PIL import Image as PILImage
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+from pipeline import ZenImageEditPipeline                        # noqa: E402
+
+HERE = os.path.dirname(os.path.abspath(__file__))
+
+
+def main():
+    ap = argparse.ArgumentParser(description="Qwen-Image-2.1 with Qwen3.5-0.8B and the adapter inside the DiT")
+    ap.add_argument("--prompt", required=True)
+    ap.add_argument("--image", nargs="*", default=[],
+                    help="condition images, order = <image1>, <image2>, ...")
+    ap.add_argument("--out", default="out.png")
+    ap.add_argument("--model", default=HERE, help="model folder (the layout shipped in this repo)")
+    ap.add_argument("--size", type=int, default=1024, help="output_resolution (frame side)")
+    ap.add_argument("--steps", type=int, default=30)
+    ap.add_argument("--seed", type=int, default=1234)
+    ap.add_argument("--device", default="cuda")
+    ap.add_argument("--no-offload", action="store_true",
+                    help="keep every component on the device (needs a large GPU)")
+    args = ap.parse_args()
+
+    pipe = ZenImageEditPipeline.from_pretrained(args.model, dtype=torch.float16)
+    pipe.set_progress_bar_config(disable=True)
+    # Phase-by-phase offload by default: the 14.5 GB fp16 DiT and the fp32 VAE decoder do not fit
+    # an 32 GB card at the same time. Keeping everything resident needs roughly 40 GB.
+    if args.device.startswith("cuda") and not args.no_offload:
+        pipe.enable_model_cpu_offload(device=args.device)
+    else:
+        pipe.to(args.device)
+
+    generator = torch.Generator(args.device).manual_seed(args.seed)
+    condition = [PILImage.open(path) for path in args.image] or None
+    if condition and len(condition) > 1 and "<image" not in args.prompt:
+        print("WARNING: with N>1 the prompt must reference <image1>, <image2>, ...", flush=True)
+    image = pipe(prompt=args.prompt,
+                 image=condition,
+                 output_resolution=args.size,
+                 num_inference_steps=args.steps,
+                 true_cfg_scale=1.0,
+                 generator=generator,
+                 output_type="pil").images[0]
+    image.save(args.out)
+    print(f"{args.size}px, {args.steps} steps, seed {args.seed} -> {args.out} {image.size}", flush=True)
+
+
+if __name__ == "__main__":
+    main()

model_index.json

@@ -0,0 +1,28 @@
+{
+  "_class_name": "ZenImageEditPipeline",
+  "_diffusers_version": "0.41.0.dev0",
+  "processor": [
+    "transformers",
+    "Qwen3VLProcessor"
+  ],
+  "scheduler": [
+    "diffusers",
+    "FlowMatchEulerDiscreteScheduler"
+  ],
+  "text_encoder": [
+    "transformers",
+    "Qwen3_5ForConditionalGeneration"
+  ],
+  "tokenizer": [
+    "transformers",
+    "Qwen2Tokenizer"
+  ],
+  "transformer": [
+    "diffusers",
+    "QwenImage21FusionTransformer2DModel"
+  ],
+  "vae": [
+    "diffusers",
+    "AutoencoderKLQwenImage21"
+  ]
+}

pipeline.py

@@ -0,0 +1,190 @@
+"""Inference pipeline for zen-image-edit: Qwen3.5-0.8B text encoder, adapter inside the DiT.
+
+Compared to `QwenImage21Pipeline` exactly one method differs: `_get_qwen_prompt_embeds` returns the
+Qwen3.5-0.8B hidden-state stack `(B, L, K*d)` plus masks instead of `(B, L, 4096)` from Qwen3-VL-8B.
+The transformer's own `text_fusion` turns that into the condition, and it also crops the system
+prompt — the adapter was trained on the full sequence (`drop_first=0`), so cropping before the
+fusion would shift every position.
+
+Latents, VAE, scheduler, edit geometry (`<imageN>` blocks, one vision slot per 2x2 latent group) and
+the prefix KV cache are inherited unchanged.
+"""
+import os
+
+import torch
+from PIL import Image as PILImage
+from diffusers.image_processor import VaeImageProcessor
+from diffusers.pipelines.qwenimage21.pipeline_qwenimage21 import QwenImage21Pipeline
+
+from transformer import QwenImage21FusionTransformer2DModel
+
+
+class ZenImageEditPipeline(QwenImage21Pipeline):
+    """Qwen-Image-2.1 with Qwen3.5-0.8B instead of Qwen3-VL-8B; one class for both modes.
+
+    Like `QwenImage21Pipeline` itself, the mode is picked by the arguments: `image=None` is
+    text-to-image, `image=[...]` is editing.
+
+    Components: `transformer` — `QwenImage21FusionTransformer2DModel` (stock DiT + `text_fusion`),
+    `text_encoder` — `Qwen3_5ForConditionalGeneration`, `processor` — the student processor
+    (images and tokenization), `tokenizer` — the student tokenizer, `vae`/`scheduler` — the
+    Qwen-Image-2.1 ones.
+    """
+
+    def __init__(
+        self,
+        scheduler,
+        vae,
+        text_encoder,
+        processor,
+        transformer,
+        tokenizer=None,
+    ):
+        # `QwenImage21Pipeline.__init__` is not reused: it derives `_drop_idx` from
+        # `processor.apply_chat_template(system)`, and the Qwen3.5 processor raises
+        # "No user query found" on a system-only message. The rest is its own code.
+        super(QwenImage21Pipeline, self).__init__()
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            processor=processor,
+            tokenizer=tokenizer,
+            transformer=transformer,
+            scheduler=scheduler,
+        )
+        self.vae_scale_factor = 16
+        self.latent_channels = self.vae.config.z_dim if getattr(self, "vae", None) else 64
+        self.image_processor = VaeImageProcessor(
+            vae_scale_factor=self.vae_scale_factor, vae_latent_channels=self.latent_channels
+        )
+        self.sys_prompt = "Comprehend and analyze the provided prompt."
+        # The prompt is a raw template string, not `apply_chat_template`: the checkpoint was
+        # trained on this one.
+        self.prompt_template_t2i = (
+            f"<|im_start|>system\n{self.sys_prompt}<|im_end|>\n"
+            f"<|im_start|>user\n{{}}<|im_end|>\n"
+            f"<|im_start|>assistant\n"
+        )
+        self.prompt_template_ti2i = (
+            f"<|im_start|>system\n{self.sys_prompt}<|im_end|>\n"
+            f"<|im_start|>user\n<image1><|vision_start|><|image_pad|><|vision_end|>{{}}<|im_end|>\n"
+            f"<|im_start|>assistant\n"
+        )
+        fusion = dict(getattr(transformer.config, "text_fusion_config", {}) or {})
+        self.student_layers = [int(i) for i in fusion.get("student_layers", ())]
+        if not self.student_layers:
+            raise ValueError("transformer config has no text_fusion_config.student_layers")
+        self._drop_idx = int(fusion.get("drop_idx", 0))
+        tokenizer = tokenizer if tokenizer is not None else getattr(processor, "tokenizer", None)
+        self._img_token_id = int(tokenizer.encode("<|image_pad|>", add_special_tokens=False)[0])
+        # The prefix crop happens inside the transformer; if the tokenizer and the config ever
+        # disagree the condition silently slides by a few positions. Check once, at build time.
+        prefix = f"<|im_start|>system\n{self.sys_prompt}<|im_end|>\n"
+        got = len(tokenizer(prefix, add_special_tokens=False)["input_ids"])
+        if got != self._drop_idx:
+            raise ValueError(
+                f"text_fusion_config.drop_idx={self._drop_idx}, but the tokenizer's system prefix "
+                f"is {got} tokens"
+            )
+
+    def _get_qwen_prompt_embeds(self, prompt=None, image=None, device=None):
+        """Student slices -> flat `(B, L, K*d)`; masks are full-length, the prefix is not cropped here."""
+        device = device or self._execution_device
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        # Qwen has no BOS token, so an empty string would leave the encoder with nothing to read.
+        prompt = [" " if not p else p for p in prompt]
+        is_t2i = image is None
+
+        if is_t2i:
+            prompts = [self.prompt_template_t2i.format(t) for t in prompt]
+            condition_pil_list = []
+        else:
+            replace = "<image1><|vision_start|><|image_pad|><|vision_end|>"
+            for i in range(2, len(image) + 1):
+                replace += f" <image{i}><|vision_start|><|image_pad|><|vision_end|>"
+            template = self.prompt_template_ti2i.replace(
+                "<image1><|vision_start|><|image_pad|><|vision_end|>", replace
+            )
+            prompts = [template.format(t) for t in prompt]
+            condition_pil_list = []
+            for _ in prompt:
+                for img in image:
+                    if not isinstance(img, PILImage.Image):
+                        img = PILImage.fromarray(img)
+                    if img.mode == "RGBA":
+                        # The checkpoint saw alpha composited over white. Encoder only: the VAE
+                        # still reads all four channels.
+                        white = PILImage.new("RGB", img.size, (255, 255, 255))
+                        white.paste(img, mask=img.getchannel("A"))
+                        img = white
+                    condition_pil_list.append(img)
+
+        # Right padding, as in adapter training: positions of the real tokens do not shift, and the
+        # fusion attention branch indexes its position table by absolute index.
+        processor_kwargs = {
+            "text": prompts,
+            "padding": True,
+            "padding_side": "right",
+            "return_tensors": "pt",
+        }
+        if not is_t2i:
+            processor_kwargs["images"] = condition_pil_list
+        model_inputs = self.processor(**processor_kwargs).to(device)
+
+        forward_kwargs = {
+            "input_ids": model_inputs.input_ids,
+            "attention_mask": model_inputs.attention_mask,
+            "output_hidden_states": True,
+        }
+        if not is_t2i and hasattr(model_inputs, "pixel_values"):
+            forward_kwargs["pixel_values"] = model_inputs.pixel_values.to(self.text_encoder.dtype)
+            forward_kwargs["image_grid_thw"] = model_inputs.image_grid_thw
+        if hasattr(model_inputs, "mm_token_type_ids"):
+            forward_kwargs["mm_token_type_ids"] = model_inputs.mm_token_type_ids
+
+        outputs = self.text_encoder(**forward_kwargs)
+        hidden = outputs.hidden_states
+        # (B, K, L, d) -> (B, L, K*d): exactly the flat input the adapter was trained on.
+        x = torch.stack([hidden[i] for i in self.student_layers], dim=1)
+        x = x.permute(0, 2, 1, 3).reshape(x.shape[0], x.shape[2], -1)
+
+        return x, model_inputs.attention_mask.bool(), (model_inputs.input_ids == self._img_token_id)
+
+    def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator):
+        """The VAE is fp32 while condition latents must be in the model dtype (fp16).
+
+        Without the round trip, `torch.cat([input_images_latents, latents])` in `__call__` promotes
+        the input to fp32 and the DiT fails on a dtype mismatch. The encoder itself runs in fp32
+        (more accurate), the result is handed back in fp16.
+        """
+        latents = super()._encode_vae_image(image.to(self.vae.dtype), generator)
+        return latents.to(next(self.transformer.parameters()).dtype)
+
+    @classmethod
+    def from_pretrained(cls, root=".", dtype=torch.float16, **kwargs):
+        """Load the pipeline from a model folder (the layout shipped in this repo)."""
+        from diffusers import AutoencoderKLQwenImage21, FlowMatchEulerDiscreteScheduler
+        from transformers import AutoProcessor, AutoTokenizer, Qwen3_5ForConditionalGeneration
+
+        if not os.path.isdir(root):
+            raise ValueError(f"expected a model folder with the components, got {root!r}")
+        transformer = QwenImage21FusionTransformer2DModel.from_pretrained(
+            os.path.join(root, "transformer"), torch_dtype=dtype
+        )
+        text_encoder = Qwen3_5ForConditionalGeneration.from_pretrained(
+            os.path.join(root, "text_encoder"), dtype=dtype
+        )
+        # The VAE stays fp32 (1.3 GB): it is the component that misbehaves in fp16.
+        vae = AutoencoderKLQwenImage21.from_pretrained(os.path.join(root, "vae"), torch_dtype=torch.float32)
+        processor = AutoProcessor.from_pretrained(os.path.join(root, "processor"))
+        tokenizer = AutoTokenizer.from_pretrained(os.path.join(root, "tokenizer"))
+        scheduler = FlowMatchEulerDiscreteScheduler.from_pretrained(os.path.join(root, "scheduler"))
+        return cls(
+            scheduler=scheduler,
+            vae=vae,
+            text_encoder=text_encoder,
+            processor=processor,
+            transformer=transformer,
+            tokenizer=tokenizer,
+            **kwargs,
+        )

processor/chat_template.jinja

@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}

processor/preprocessor_config.json

@@ -0,0 +1,21 @@
+{
+    "size": {
+        "longest_edge": 16777216,
+        "shortest_edge": 65536
+    },
+    "patch_size": 16,
+    "temporal_patch_size": 2,
+    "merge_size": 2,
+    "image_mean": [
+        0.5,
+        0.5,
+        0.5
+    ],
+    "image_std": [
+        0.5,
+        0.5,
+        0.5
+    ],
+    "processor_class": "Qwen3VLProcessor",
+    "image_processor_type": "Qwen2VLImageProcessorFast"
+}

processor/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f9e4d4901a92b997e463c1f46055088b6cca5ca61a6522d1b9f64c4bb81cb42
+size 12807982

processor/tokenizer_config.json

@@ -0,0 +1,305 @@
+{
+    "add_prefix_space": false,
+    "added_tokens_decoder": {
+        "248044": {
+            "content": "<|endoftext|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248045": {
+            "content": "<|im_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248046": {
+            "content": "<|im_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248047": {
+            "content": "<|object_ref_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248048": {
+            "content": "<|object_ref_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248049": {
+            "content": "<|box_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248050": {
+            "content": "<|box_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248051": {
+            "content": "<|quad_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248052": {
+            "content": "<|quad_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248053": {
+            "content": "<|vision_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248054": {
+            "content": "<|vision_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248055": {
+            "content": "<|vision_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248056": {
+            "content": "<|image_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248057": {
+            "content": "<|video_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248058": {
+            "content": "<tool_call>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248059": {
+            "content": "</tool_call>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248060": {
+            "content": "<|fim_prefix|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248061": {
+            "content": "<|fim_middle|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248062": {
+            "content": "<|fim_suffix|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248063": {
+            "content": "<|fim_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248064": {
+            "content": "<|repo_name|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248065": {
+            "content": "<|file_sep|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248066": {
+            "content": "<tool_response>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248067": {
+            "content": "</tool_response>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248068": {
+            "content": "<think>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248069": {
+            "content": "</think>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248070": {
+            "content": "<|audio_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248071": {
+            "content": "<|audio_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248072": {
+            "content": "<tts_pad>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248073": {
+            "content": "<tts_text_bos>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248074": {
+            "content": "<tts_text_eod>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248075": {
+            "content": "<tts_text_bos_single>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248076": {
+            "content": "<|audio_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        }
+    },
+    "additional_special_tokens": [
+        "<|im_start|>",
+        "<|im_end|>",
+        "<|object_ref_start|>",
+        "<|object_ref_end|>",
+        "<|box_start|>",
+        "<|box_end|>",
+        "<|quad_start|>",
+        "<|quad_end|>",
+        "<|vision_start|>",
+        "<|vision_end|>",
+        "<|vision_pad|>",
+        "<|image_pad|>",
+        "<|video_pad|>"
+    ],
+    "bos_token": null,
+    "chat_template": "{%- set image_count = namespace(value=0) %}\n{%- set video_count = namespace(value=0) %}\n{%- macro render_content(content, do_vision_count, is_system_content=false) %}\n    {%- if content is string %}\n        {{- content }}\n    {%- elif content is iterable and content is not mapping %}\n        {%- for item in content %}\n            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}\n                {%- if is_system_content %}\n                    {{- raise_exception('System message cannot contain images.') }}\n                {%- endif %}\n                {%- if do_vision_count %}\n                    {%- set image_count.value = image_count.value + 1 %}\n                {%- endif %}\n                {%- if add_vision_id %}\n                    {{- 'Picture ' ~ image_count.value ~ ': ' }}\n                {%- endif %}\n                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}\n            {%- elif 'video' in item or item.type == 'video' %}\n                {%- if is_system_content %}\n                    {{- raise_exception('System message cannot contain videos.') }}\n                {%- endif %}\n                {%- if do_vision_count %}\n                    {%- set video_count.value = video_count.value + 1 %}\n                {%- endif %}\n                {%- if add_vision_id %}\n                    {{- 'Video ' ~ video_count.value ~ ': ' }}\n                {%- endif %}\n                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}\n            {%- elif 'text' in item %}\n                {{- item.text }}\n            {%- else %}\n                {{- raise_exception('Unexpected item type in content.') }}\n            {%- endif %}\n        {%- endfor %}\n    {%- elif content is none or content is undefined %}\n        {{- '' }}\n    {%- else %}\n        {{- raise_exception('Unexpected content type.') }}\n    {%- endif %}\n{%- endmacro %}\n{%- if not messages %}\n    {{- raise_exception('No messages provided.') }}\n{%- endif %}\n{%- if tools and tools is iterable and tools is not mapping %}\n    {{- '<|im_start|>system\\n' }}\n    {{- \"# Tools\\n\\nYou have access to the following functions:\\n\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\" }}\n    {{- '\\n\\nIf you choose to call a function ONLY reply in the following format with NO suffix:\\n\\n<tool_call>\\n<function=example_function_name>\\n<parameter=example_parameter_1>\\nvalue_1\\n</parameter>\\n<parameter=example_parameter_2>\\nThis is the value for the second parameter\\nthat can span\\nmultiple lines\\n</parameter>\\n</function>\\n</tool_call>\\n\\n<IMPORTANT>\\nReminder:\\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\\n- Required parameters MUST be specified\\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\\n</IMPORTANT>' }}\n    {%- if messages[0].role == 'system' %}\n        {%- set content = render_content(messages[0].content, false, true)|trim %}\n        {%- if content %}\n            {{- '\\n\\n' + content }}\n        {%- endif %}\n    {%- endif %}\n    {{- '<|im_end|>\\n' }}\n{%- else %}\n    {%- if messages[0].role == 'system' %}\n        {%- set content = render_content(messages[0].content, false, true)|trim %}\n        {{- '<|im_start|>system\\n' + content + '<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n    {%- set index = (messages|length - 1) - loop.index0 %}\n    {%- if ns.multi_step_tool and message.role == \"user\" %}\n        {%- set content = render_content(message.content, false)|trim %}\n        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}\n            {%- set ns.multi_step_tool = false %}\n            {%- set ns.last_query_index = index %}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if ns.multi_step_tool %}\n    {{- raise_exception('No user query found in messages.') }}\n{%- endif %}\n{%- for message in messages %}\n    {%- set content = render_content(message.content, true)|trim %}\n    {%- if message.role == \"system\" %}\n        {%- if not loop.first %}\n            {{- raise_exception('System message must be at the beginning.') }}\n        {%- endif %}\n    {%- elif message.role == \"user\" %}\n        {{- '<|im_start|>' + message.role + '\\n' + content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {%- set reasoning_content = '' %}\n        {%- if message.reasoning_content is string %}\n            {%- set reasoning_content = message.reasoning_content %}\n        {%- else %}\n            {%- if '</think>' in content %}\n                {%- set reasoning_content = content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n                {%- set content = content.split('</think>')[-1].lstrip('\\n') %}\n            {%- endif %}\n        {%- endif %}\n        {%- set reasoning_content = reasoning_content|trim %}\n        {%- if loop.index0 > ns.last_query_index %}\n            {{- '<|im_start|>' + message.role + '\\n<think>\\n' + reasoning_content + '\\n</think>\\n\\n' + content }}\n        {%- else %}\n            {{- '<|im_start|>' + message.role + '\\n' + content }}\n        {%- endif %}\n        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}\n            {%- for tool_call in message.tool_calls %}\n                {%- if tool_call.function is defined %}\n                    {%- set tool_call = tool_call.function %}\n                {%- endif %}\n                {%- if loop.first %}\n                    {%- if content|trim %}\n                        {{- '\\n\\n<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                    {%- else %}\n                        {{- '<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                    {%- endif %}\n                {%- else %}\n                    {{- '\\n<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                {%- endif %}\n                {%- if tool_call.arguments is defined %}\n                    {%- for args_name, args_value in tool_call.arguments|items %}\n                        {{- '<parameter=' + args_name + '>\\n' }}\n                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}\n                        {{- args_value }}\n                        {{- '\\n</parameter>\\n' }}\n                    {%- endfor %}\n                {%- endif %}\n                {{- '</function>\\n</tool_call>' }}\n            {%- endfor %}\n        {%- endif %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if loop.previtem and loop.previtem.role != \"tool\" %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- content }}\n        {{- '\\n</tool_response>' }}\n        {%- if not loop.last and loop.nextitem.role != \"tool\" %}\n            {{- '<|im_end|>\\n' }}\n        {%- elif loop.last %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- else %}\n        {{- raise_exception('Unexpected message role.') }}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n    {%- if enable_thinking is defined and enable_thinking is true %}\n        {{- '<think>\\n' }}\n    {%- else %}\n        {{- '<think>\\n\\n</think>\\n\\n' }}\n    {%- endif %}\n{%- endif %}",
+    "clean_up_tokenization_spaces": false,
+    "eos_token": "<|im_end|>",
+    "errors": "replace",
+    "model_max_length": 262144,
+    "pad_token": "<|endoftext|>",
+    "split_special_tokens": false,
+    "tokenizer_class": "Qwen2Tokenizer",
+    "unk_token": null,
+    "add_bos_token": false,
+    "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+    "extra_special_tokens": {
+        "audio_bos_token": "<|audio_start|>",
+        "audio_eos_token": "<|audio_end|>",
+        "audio_token": "<|audio_pad|>",
+        "image_token": "<|image_pad|>",
+        "video_token": "<|video_pad|>",
+        "vision_bos_token": "<|vision_start|>",
+        "vision_eos_token": "<|vision_end|>"
+    }
+}

processor/video_preprocessor_config.json

@@ -0,0 +1,21 @@
+{
+    "size": {
+        "longest_edge": 25165824,
+        "shortest_edge": 4096
+    },
+    "patch_size": 16,
+    "temporal_patch_size": 2,
+    "merge_size": 2,
+    "image_mean": [
+        0.5,
+        0.5,
+        0.5
+    ],
+    "image_std": [
+        0.5,
+        0.5,
+        0.5
+    ],
+    "processor_class": "Qwen3VLProcessor",
+    "video_processor_type": "Qwen3VLVideoProcessor"
+}

scheduler/scheduler_config.json

@@ -0,0 +1,18 @@
+{
+  "_class_name": "FlowMatchEulerDiscreteScheduler",
+  "_diffusers_version": "0.37.0.dev0",
+  "base_image_seq_len": 256,
+  "base_shift": 0.5,
+  "invert_sigmas": false,
+  "max_image_seq_len": 8192,
+  "max_shift": 0.9,
+  "num_train_timesteps": 1000,
+  "shift": 1.0,
+  "shift_terminal": 0.02,
+  "stochastic_sampling": false,
+  "time_shift_type": "exponential",
+  "use_beta_sigmas": false,
+  "use_dynamic_shifting": true,
+  "use_exponential_sigmas": false,
+  "use_karras_sigmas": false
+}

text_encoder/config.json

@@ -0,0 +1,96 @@
+{
+    "architectures": [
+        "Qwen3_5ForConditionalGeneration"
+    ],
+    "image_token_id": 248056,
+    "model_type": "qwen3_5",
+    "text_config": {
+        "attention_bias": false,
+        "attention_dropout": 0.0,
+        "attn_output_gate": true,
+        "dtype": "bfloat16",
+        "eos_token_id": 248044,
+        "full_attention_interval": 4,
+        "head_dim": 256,
+        "hidden_act": "silu",
+        "hidden_size": 1024,
+        "initializer_range": 0.02,
+        "intermediate_size": 3584,
+        "layer_types": [
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention"
+        ],
+        "linear_conv_kernel_dim": 4,
+        "linear_key_head_dim": 128,
+        "linear_num_key_heads": 16,
+        "linear_num_value_heads": 16,
+        "linear_value_head_dim": 128,
+        "max_position_embeddings": 262144,
+        "mlp_only_layers": [],
+        "model_type": "qwen3_5_text",
+        "mtp_num_hidden_layers": 1,
+        "mtp_use_dedicated_embeddings": false,
+        "num_attention_heads": 8,
+        "num_hidden_layers": 24,
+        "num_key_value_heads": 2,
+        "rms_norm_eps": 1e-06,
+        "tie_word_embeddings": true,
+        "use_cache": true,
+        "vocab_size": 248320,
+        "mamba_ssm_dtype": "float32",
+        "rope_parameters": {
+            "mrope_interleaved": true,
+            "mrope_section": [
+                11,
+                11,
+                10
+            ],
+            "rope_type": "default",
+            "rope_theta": 10000000,
+            "partial_rotary_factor": 0.25
+        }
+    },
+    "tie_word_embeddings": true,
+    "transformers_version": "4.57.0.dev0",
+    "video_token_id": 248057,
+    "vision_config": {
+        "deepstack_visual_indexes": [],
+        "depth": 12,
+        "hidden_act": "gelu_pytorch_tanh",
+        "hidden_size": 768,
+        "in_channels": 3,
+        "initializer_range": 0.02,
+        "intermediate_size": 3072,
+        "model_type": "qwen3_5",
+        "num_heads": 12,
+        "num_position_embeddings": 2304,
+        "out_hidden_size": 1024,
+        "patch_size": 16,
+        "spatial_merge_size": 2,
+        "temporal_patch_size": 2
+    },
+    "vision_end_token_id": 248054,
+    "vision_start_token_id": 248053
+}

text_encoder/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:22978c1272be438dc4a22dbaf3ce1f6843d1ee2dc67490387d1afc8c88384943
+size 1746937368

tokenizer/chat_template.jinja

@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}

tokenizer/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f9e4d4901a92b997e463c1f46055088b6cca5ca61a6522d1b9f64c4bb81cb42
+size 12807982

tokenizer/tokenizer_config.json

@@ -0,0 +1,305 @@
+{
+    "add_prefix_space": false,
+    "added_tokens_decoder": {
+        "248044": {
+            "content": "<|endoftext|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248045": {
+            "content": "<|im_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248046": {
+            "content": "<|im_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248047": {
+            "content": "<|object_ref_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248048": {
+            "content": "<|object_ref_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248049": {
+            "content": "<|box_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248050": {
+            "content": "<|box_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248051": {
+            "content": "<|quad_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248052": {
+            "content": "<|quad_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248053": {
+            "content": "<|vision_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248054": {
+            "content": "<|vision_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248055": {
+            "content": "<|vision_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248056": {
+            "content": "<|image_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248057": {
+            "content": "<|video_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248058": {
+            "content": "<tool_call>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248059": {
+            "content": "</tool_call>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248060": {
+            "content": "<|fim_prefix|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248061": {
+            "content": "<|fim_middle|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248062": {
+            "content": "<|fim_suffix|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248063": {
+            "content": "<|fim_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248064": {
+            "content": "<|repo_name|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248065": {
+            "content": "<|file_sep|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248066": {
+            "content": "<tool_response>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248067": {
+            "content": "</tool_response>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248068": {
+            "content": "<think>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248069": {
+            "content": "</think>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248070": {
+            "content": "<|audio_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248071": {
+            "content": "<|audio_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248072": {
+            "content": "<tts_pad>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248073": {
+            "content": "<tts_text_bos>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248074": {
+            "content": "<tts_text_eod>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248075": {
+            "content": "<tts_text_bos_single>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248076": {
+            "content": "<|audio_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        }
+    },
+    "additional_special_tokens": [
+        "<|im_start|>",
+        "<|im_end|>",
+        "<|object_ref_start|>",
+        "<|object_ref_end|>",
+        "<|box_start|>",
+        "<|box_end|>",
+        "<|quad_start|>",
+        "<|quad_end|>",
+        "<|vision_start|>",
+        "<|vision_end|>",
+        "<|vision_pad|>",
+        "<|image_pad|>",
+        "<|video_pad|>"
+    ],
+    "bos_token": null,
+    "chat_template": "{%- set image_count = namespace(value=0) %}\n{%- set video_count = namespace(value=0) %}\n{%- macro render_content(content, do_vision_count, is_system_content=false) %}\n    {%- if content is string %}\n        {{- content }}\n    {%- elif content is iterable and content is not mapping %}\n        {%- for item in content %}\n            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}\n                {%- if is_system_content %}\n                    {{- raise_exception('System message cannot contain images.') }}\n                {%- endif %}\n                {%- if do_vision_count %}\n                    {%- set image_count.value = image_count.value + 1 %}\n                {%- endif %}\n                {%- if add_vision_id %}\n                    {{- 'Picture ' ~ image_count.value ~ ': ' }}\n                {%- endif %}\n                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}\n            {%- elif 'video' in item or item.type == 'video' %}\n                {%- if is_system_content %}\n                    {{- raise_exception('System message cannot contain videos.') }}\n                {%- endif %}\n                {%- if do_vision_count %}\n                    {%- set video_count.value = video_count.value + 1 %}\n                {%- endif %}\n                {%- if add_vision_id %}\n                    {{- 'Video ' ~ video_count.value ~ ': ' }}\n                {%- endif %}\n                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}\n            {%- elif 'text' in item %}\n                {{- item.text }}\n            {%- else %}\n                {{- raise_exception('Unexpected item type in content.') }}\n            {%- endif %}\n        {%- endfor %}\n    {%- elif content is none or content is undefined %}\n        {{- '' }}\n    {%- else %}\n        {{- raise_exception('Unexpected content type.') }}\n    {%- endif %}\n{%- endmacro %}\n{%- if not messages %}\n    {{- raise_exception('No messages provided.') }}\n{%- endif %}\n{%- if tools and tools is iterable and tools is not mapping %}\n    {{- '<|im_start|>system\\n' }}\n    {{- \"# Tools\\n\\nYou have access to the following functions:\\n\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\" }}\n    {{- '\\n\\nIf you choose to call a function ONLY reply in the following format with NO suffix:\\n\\n<tool_call>\\n<function=example_function_name>\\n<parameter=example_parameter_1>\\nvalue_1\\n</parameter>\\n<parameter=example_parameter_2>\\nThis is the value for the second parameter\\nthat can span\\nmultiple lines\\n</parameter>\\n</function>\\n</tool_call>\\n\\n<IMPORTANT>\\nReminder:\\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\\n- Required parameters MUST be specified\\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\\n</IMPORTANT>' }}\n    {%- if messages[0].role == 'system' %}\n        {%- set content = render_content(messages[0].content, false, true)|trim %}\n        {%- if content %}\n            {{- '\\n\\n' + content }}\n        {%- endif %}\n    {%- endif %}\n    {{- '<|im_end|>\\n' }}\n{%- else %}\n    {%- if messages[0].role == 'system' %}\n        {%- set content = render_content(messages[0].content, false, true)|trim %}\n        {{- '<|im_start|>system\\n' + content + '<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n    {%- set index = (messages|length - 1) - loop.index0 %}\n    {%- if ns.multi_step_tool and message.role == \"user\" %}\n        {%- set content = render_content(message.content, false)|trim %}\n        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}\n            {%- set ns.multi_step_tool = false %}\n            {%- set ns.last_query_index = index %}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if ns.multi_step_tool %}\n    {{- raise_exception('No user query found in messages.') }}\n{%- endif %}\n{%- for message in messages %}\n    {%- set content = render_content(message.content, true)|trim %}\n    {%- if message.role == \"system\" %}\n        {%- if not loop.first %}\n            {{- raise_exception('System message must be at the beginning.') }}\n        {%- endif %}\n    {%- elif message.role == \"user\" %}\n        {{- '<|im_start|>' + message.role + '\\n' + content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {%- set reasoning_content = '' %}\n        {%- if message.reasoning_content is string %}\n            {%- set reasoning_content = message.reasoning_content %}\n        {%- else %}\n            {%- if '</think>' in content %}\n                {%- set reasoning_content = content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n                {%- set content = content.split('</think>')[-1].lstrip('\\n') %}\n            {%- endif %}\n        {%- endif %}\n        {%- set reasoning_content = reasoning_content|trim %}\n        {%- if loop.index0 > ns.last_query_index %}\n            {{- '<|im_start|>' + message.role + '\\n<think>\\n' + reasoning_content + '\\n</think>\\n\\n' + content }}\n        {%- else %}\n            {{- '<|im_start|>' + message.role + '\\n' + content }}\n        {%- endif %}\n        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}\n            {%- for tool_call in message.tool_calls %}\n                {%- if tool_call.function is defined %}\n                    {%- set tool_call = tool_call.function %}\n                {%- endif %}\n                {%- if loop.first %}\n                    {%- if content|trim %}\n                        {{- '\\n\\n<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                    {%- else %}\n                        {{- '<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                    {%- endif %}\n                {%- else %}\n                    {{- '\\n<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                {%- endif %}\n                {%- if tool_call.arguments is defined %}\n                    {%- for args_name, args_value in tool_call.arguments|items %}\n                        {{- '<parameter=' + args_name + '>\\n' }}\n                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}\n                        {{- args_value }}\n                        {{- '\\n</parameter>\\n' }}\n                    {%- endfor %}\n                {%- endif %}\n                {{- '</function>\\n</tool_call>' }}\n            {%- endfor %}\n        {%- endif %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if loop.previtem and loop.previtem.role != \"tool\" %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- content }}\n        {{- '\\n</tool_response>' }}\n        {%- if not loop.last and loop.nextitem.role != \"tool\" %}\n            {{- '<|im_end|>\\n' }}\n        {%- elif loop.last %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- else %}\n        {{- raise_exception('Unexpected message role.') }}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n    {%- if enable_thinking is defined and enable_thinking is true %}\n        {{- '<think>\\n' }}\n    {%- else %}\n        {{- '<think>\\n\\n</think>\\n\\n' }}\n    {%- endif %}\n{%- endif %}",
+    "clean_up_tokenization_spaces": false,
+    "eos_token": "<|im_end|>",
+    "errors": "replace",
+    "model_max_length": 262144,
+    "pad_token": "<|endoftext|>",
+    "split_special_tokens": false,
+    "tokenizer_class": "Qwen2Tokenizer",
+    "unk_token": null,
+    "add_bos_token": false,
+    "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+    "extra_special_tokens": {
+        "audio_bos_token": "<|audio_start|>",
+        "audio_eos_token": "<|audio_end|>",
+        "audio_token": "<|audio_pad|>",
+        "image_token": "<|image_pad|>",
+        "video_token": "<|video_pad|>",
+        "vision_bos_token": "<|vision_start|>",
+        "vision_eos_token": "<|vision_end|>"
+    }
+}

transformer.py

@@ -0,0 +1,278 @@
+"""Qwen-Image-2.1 DiT with the text adapter built in: `text_fusion` right before `txt_in`.
+
+The adapter (`image21-08b-text-encoder-adapter` v11) is a regression from Qwen3.5-0.8B hidden
+states to the input of the native `txt_in = QwenImage21TextProjection`. It therefore sits
+*before* `txt_in`, exactly like `Krea2TextFusion` sits before `Krea2TextProjection` in Krea 2:
+
+    student slices (B, L, K*d) -> text_fusion -> (B, L, 4096) -> txt_in -> joint stream
+
+    y = MLP(x) + Attn(x) + Mixer(x)
+        |         |         +-- attention over the slice axis (K student layers)
+        |         +------------ self-attention over tokens, key padding mask
+        +---------------------- ln-per-layer over slices (norms grow ~40x with depth)
+
+Two details matter and are easy to get wrong:
+
+* The 14 system-prompt tokens are sliced off **after** the fusion, not before. The adapter was
+  trained on the full sequence (`drop_first=0`) and its attention branch indexes a learned
+  position table by absolute token index, so cropping first would shift every position.
+* The config key is named `text_fusion_config`, not `text_fusion`: `ModelMixin.__getattr__`
+  returns a config value before `nn.Module` can hand back a submodule, so a same-named key made
+  `model.text_fusion` a `dict` and weight loading failed.
+"""
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from diffusers.models.transformers.transformer_qwenimage21 import (
+    QwenImage21Transformer2DModel,
+)
+
+
+class PerLayerNorm(nn.Module):
+    """Normalize every slice separately, then the concatenation.
+
+    Without per-slice normalization the early layers barely reach the output: hidden-state norms
+    grow roughly 40x from layer 2 to layer 27.
+    """
+
+    def __init__(self, in_dim, n_slices):
+        super().__init__()
+        assert in_dim % n_slices == 0, f"{in_dim} is not divisible by {n_slices}"
+        self.n, self.d = n_slices, in_dim // n_slices
+        self.per = nn.LayerNorm(self.d, elementwise_affine=False)
+        self.all = nn.LayerNorm(in_dim)
+
+    def forward(self, x):
+        b, l, _ = x.shape
+        return self.all(self.per(x.view(b, l, self.n, self.d)).reshape(b, l, -1))
+
+
+class AttnBlock(nn.Module):
+    """Pre-norm self-attention + FFN. No causality: text is not autoregressive and the DiT
+    already sees the whole sequence at once."""
+
+    def __init__(self, d_model, n_heads, ffn_mult=4):
+        super().__init__()
+        self.n1 = nn.LayerNorm(d_model)
+        self.attn = nn.MultiheadAttention(d_model, n_heads, batch_first=True)
+        self.n2 = nn.LayerNorm(d_model)
+        self.ffn = nn.Sequential(nn.Linear(d_model, d_model * ffn_mult),
+                                 nn.GELU(approximate="tanh"),
+                                 nn.Linear(d_model * ffn_mult, d_model))
+
+    def forward(self, x, key_padding_mask=None):
+        h = self.n1(x)
+        x = x + self.attn(h, h, h, need_weights=False, key_padding_mask=key_padding_mask)[0]
+        return x + self.ffn(self.n2(x))
+
+
+class AttnMixer(nn.Module):
+    """Token branch: norm -> project to d_model -> N blocks -> project to out_dim.
+
+    The last projection is zero-initialized, so at step 0 the branch adds nothing.
+    """
+
+    def __init__(self, in_dim, out_dim, d_model=1024, n_heads=8, blocks=2, max_len=256):
+        super().__init__()
+        self.norm = nn.LayerNorm(in_dim)
+        self.inp = nn.Linear(in_dim, d_model)
+        self.pos = nn.Parameter(torch.zeros(1, max_len, d_model))
+        self.blocks = nn.ModuleList([AttnBlock(d_model, n_heads) for _ in range(blocks)])
+        self.out = nn.Linear(d_model, out_dim)
+        nn.init.zeros_(self.out.weight)
+        nn.init.zeros_(self.out.bias)
+
+    def forward(self, x, key_padding_mask=None):
+        h = self.inp(self.norm(x))
+        n = h.shape[1]
+        pos = self.pos
+        if n > pos.shape[1]:
+            # The position table was trained on `max_len` slots; the tail is padded with zeros
+            # (long sequences simply carry no positional signal there, but nothing crashes).
+            pos = F.pad(pos, (0, 0, 0, n - pos.shape[1]))
+        h = h + pos[:, :n]
+        for block in self.blocks:
+            h = block(h, key_padding_mask)
+        return self.out(h)
+
+
+class SliceMixer(nn.Module):
+    """Attention over the slice axis (the `layerwise_blocks` + `projector` part of Krea 2 fusion).
+
+    Slices are normalized one by one and run as a sequence of length K through N self-attention
+    blocks; `Linear(K -> 1)` then collapses the axis. The result is added to the MLP output
+    (zero-initialized output projection).
+    """
+
+    def __init__(self, n_slices, d, out_dim, n_heads=8, blocks=2, ffn_mult=2):
+        super().__init__()
+        self.n, self.d = n_slices, d
+        self.per = nn.LayerNorm(d, elementwise_affine=False)
+        self.blocks = nn.ModuleList([AttnBlock(d, n_heads, ffn_mult) for _ in range(blocks)])
+        self.proj = nn.Linear(n_slices, 1, bias=False)
+        self.out = nn.Linear(d, out_dim)
+        nn.init.zeros_(self.out.weight)
+        nn.init.zeros_(self.out.bias)
+
+    def forward(self, x):
+        b, l, _ = x.shape
+        h = self.per(x.view(b, l, self.n, self.d)).reshape(b * l, self.n, self.d)
+        for block in self.blocks:
+            h = block(h)
+        h = self.proj(h.permute(0, 2, 1)).squeeze(-1)
+        return self.out(h).reshape(b, l, -1)
+
+
+class AttnAdapter(nn.Module):
+    """Point-wise MLP (`self.mlp`) plus residual branches over tokens (`self.attn`) and slices (`self.mixer`)."""
+
+    def __init__(self, mods, attn=None, mixer=None):
+        super().__init__()
+        self.mlp = nn.Sequential(*mods)
+        self.attn = attn
+        self.mixer = mixer
+
+    def forward(self, x, mask=None):
+        """`mask`: `(B, L)` bool, True = real token. The attention branches get `key_padding_mask = ~mask`,
+        otherwise attention would look into the padding."""
+        out = self.mlp(x)
+        if self.attn is not None:
+            kpm = None if mask is None else ~mask.bool()
+            out = out + self.attn(x, kpm)
+        if self.mixer is not None:
+            out = out + self.mixer(x)
+        return out
+
+
+def build_fusion(in_dim, out_dim, hidden=4096, proj_layers=2, norm="none", n_slices=1,
+                 attention=0, attn_dim=1024, attn_heads=8, max_len=256,
+                 mixer=0, mixer_heads=8, mixer_ffn=2, **_ignored):
+    """Build the fusion block from the transformer config (`text_fusion_config`).
+
+    `proj_layers` linear layers with GELU(tanh) between them; `attention`/`mixer` are the number of
+    residual branches. Extra config keys (`student_layers`, `drop_idx`) are ignored here: the
+    pipeline uses them, the block does not.
+    """
+    if proj_layers < 2:
+        raise ValueError("at least 2 linear layers are required")
+    mods = []
+    if norm == "ln":
+        mods.append(nn.LayerNorm(in_dim))
+    elif norm == "ln-per-layer":
+        mods.append(PerLayerNorm(in_dim, n_slices))
+    elif norm == "rms":
+        mods.append(nn.RMSNorm(in_dim))
+    elif norm != "none":
+        raise ValueError(f"unknown norm: {norm}")
+    mods += [nn.Linear(in_dim, hidden), nn.GELU(approximate="tanh")]
+    for _ in range(proj_layers - 2):
+        mods += [nn.Linear(hidden, hidden), nn.GELU(approximate="tanh")]
+    mods.append(nn.Linear(hidden, out_dim))
+    attn = AttnMixer(in_dim, out_dim, attn_dim, attn_heads, attention, max_len) if attention else None
+    mix = SliceMixer(n_slices, in_dim // n_slices, out_dim, mixer_heads, mixer, mixer_ffn) if mixer else None
+    if attn is not None or mix is not None:
+        return AttnAdapter(mods, attn, mix)
+    return nn.Sequential(*mods)
+
+
+class QwenImage21FusionTransformer2DModel(QwenImage21Transformer2DModel):
+    """`QwenImage21Transformer2DModel` + `text_fusion` (student stack -> 4096 condition).
+
+    The config carries an extra key `text_fusion_config` (arguments of `build_fusion` plus
+    `student_layers` and `drop_idx` for the pipeline), so the checkpoint is self-contained: the
+    block weights live in the same file under the `text_fusion.` prefix.
+
+    The `__init__` signature intentionally repeats the parent's. `ConfigMixin.extract_init_dict`
+    builds `init_dict` from named parameters only and ignores `**kwargs`, so forwarding the config
+    through `**kwargs` would drop the parent keys and rebuild the model from defaults.
+    """
+
+    _no_split_modules = QwenImage21Transformer2DModel._no_split_modules + ["AttnAdapter"]
+
+    def __init__(
+        self,
+        patch_size: int = 1,
+        in_channels: int = 64,
+        out_channels: int | None = 64,
+        num_layers: int = 32,
+        attention_head_dim: int = 128,
+        num_attention_heads: int = 32,
+        context_in_dim: int = 4096,
+        mlp_ratio: int = 3,
+        axes_dims_rope: tuple[int, int, int] = (16, 56, 56),
+        eps: float = 1e-6,
+        causal_condition: bool = True,
+        text_fusion_config: dict | None = None,
+    ):
+        super().__init__(
+            patch_size=patch_size,
+            in_channels=in_channels,
+            out_channels=out_channels,
+            num_layers=num_layers,
+            attention_head_dim=attention_head_dim,
+            num_attention_heads=num_attention_heads,
+            context_in_dim=context_in_dim,
+            mlp_ratio=mlp_ratio,
+            axes_dims_rope=axes_dims_rope,
+            eps=eps,
+            causal_condition=causal_condition,
+        )
+        if text_fusion_config is None:
+            raise ValueError(
+                "`text_fusion_config` is required in config.json: without it there is nowhere to "
+                "attach the adapter, and `encoder_hidden_states` are expected to be context_in_dim"
+            )
+        self.text_fusion = build_fusion(**text_fusion_config)
+        self.register_to_config(text_fusion_config=dict(text_fusion_config))
+        self.condition_drop_idx = int(text_fusion_config.get("drop_idx", 0))
+
+    @property
+    def text_fusion_dtype(self):
+        return next(self.text_fusion.parameters()).dtype
+
+    def _condition(self, encoder_hidden_states, encoder_hidden_states_mask, img_mask):
+        """Student slices -> DiT condition: fusion over the full sequence, then crop the prefix.
+
+        `img_mask` arrives from the pipeline already concatenated with the target-image slots
+        (`append_target_slots`), so only the conditioning part is cropped — otherwise the target
+        slots would slide out of place.
+        """
+        length = encoder_hidden_states.shape[1]
+        mask = None if encoder_hidden_states_mask is None else encoder_hidden_states_mask.bool()
+        fused = self.text_fusion(encoder_hidden_states.to(self.text_fusion_dtype), mask)
+        drop = self.condition_drop_idx
+        if not drop:
+            return fused, encoder_hidden_states_mask, img_mask
+        fused = fused[:, drop:]
+        if encoder_hidden_states_mask is not None:
+            encoder_hidden_states_mask = encoder_hidden_states_mask[:, drop:]
+        img_mask = torch.cat([img_mask[:, drop:length], img_mask[:, length:]], dim=1)
+        return fused, encoder_hidden_states_mask, img_mask
+
+    def forward(
+        self,
+        hidden_states,
+        encoder_hidden_states,
+        timestep,
+        img_shapes,
+        img_mask,
+        encoder_hidden_states_mask=None,
+        **kwargs,
+    ):
+        """`encoder_hidden_states` here is the student stack `(B, L, K*d)`, not a ready condition.
+
+        The fusion output is cast back to the latent dtype: the block is stored in fp16 while
+        `txt_in` and the transformer blocks expect the model dtype.
+        """
+        fused, encoder_hidden_states_mask, img_mask = self._condition(
+            encoder_hidden_states, encoder_hidden_states_mask, img_mask
+        )
+        return super().forward(
+            hidden_states=hidden_states,
+            encoder_hidden_states=fused.to(hidden_states.dtype),
+            timestep=timestep,
+            img_shapes=img_shapes,
+            img_mask=img_mask,
+            encoder_hidden_states_mask=encoder_hidden_states_mask,
+            **kwargs,
+        )

transformer/config.json

@@ -0,0 +1,43 @@
+{
+  "_class_name": "QwenImage21FusionTransformer2DModel",
+  "_diffusers_version": "0.37.0.dev0",
+  "attention_head_dim": 128,
+  "axes_dims_rope": [
+    16,
+    56,
+    56
+  ],
+  "context_in_dim": 4096,
+  "in_channels": 64,
+  "num_attention_heads": 32,
+  "num_layers": 32,
+  "out_channels": 64,
+  "patch_size": 1,
+  "mlp_ratio": 3,
+  "eps": 1e-06,
+  "causal_condition": true,
+  "text_fusion_config": {
+    "in_dim": 6144,
+    "out_dim": 4096,
+    "hidden": 6144,
+    "proj_layers": 3,
+    "norm": "ln-per-layer",
+    "n_slices": 6,
+    "attention": 2,
+    "attn_dim": 1024,
+    "attn_heads": 8,
+    "max_len": 512,
+    "mixer": 2,
+    "mixer_heads": 8,
+    "mixer_ffn": 2,
+    "student_layers": [
+      4,
+      8,
+      12,
+      16,
+      20,
+      24
+    ],
+    "drop_idx": 14
+  }
+}

transformer/diffusion_pytorch_model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a71f3c4eca68c67df8c8fd288cc1daf27ac0a89d6eee9f0ab7aa2ec1031adbe4
+size 10284159452

transformer/diffusion_pytorch_model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:05793f413d6bf507cee006142f88746901617e641b0c05648298c7ba17bd6e36
+size 4261951792

transformer/diffusion_pytorch_model.safetensors.index.json

@@ -0,0 +1,370 @@
+{
+  "metadata": {
+    "total_size": 14546069516
+  },
+  "weight_map": {
+    "img_in.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "modulation.1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.0.attn.in_proj_bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.0.attn.in_proj_weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.0.attn.out_proj.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.0.attn.out_proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.0.ffn.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.0.ffn.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.0.ffn.2.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.0.ffn.2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.0.n1.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.0.n1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.0.n2.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.0.n2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.1.attn.in_proj_bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.1.attn.in_proj_weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.1.attn.out_proj.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.1.attn.out_proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.1.ffn.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.1.ffn.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.1.ffn.2.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.1.ffn.2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.1.n1.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.1.n1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.1.n2.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.blocks.1.n2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.inp.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.inp.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.norm.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.norm.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.out.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.attn.pos": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.0.attn.in_proj_bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.0.attn.in_proj_weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.0.attn.out_proj.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.0.attn.out_proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.0.ffn.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.0.ffn.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.0.ffn.2.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.0.ffn.2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.0.n1.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.0.n1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.0.n2.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.0.n2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.1.attn.in_proj_bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.1.attn.in_proj_weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.1.attn.out_proj.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.1.attn.out_proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.1.ffn.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.1.ffn.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.1.ffn.2.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.1.ffn.2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.1.n1.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.1.n1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.1.n2.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.blocks.1.n2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.out.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mixer.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mlp.0.all.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mlp.0.all.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mlp.1.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mlp.1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mlp.3.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mlp.3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mlp.5.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "text_fusion.mlp.5.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "time_text_embed.timestep_embedder.linear_1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "time_text_embed.timestep_embedder.linear_2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.0.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.0.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.0.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.0.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.0.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.0.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.0.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.0.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.0.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.1.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.1.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.1.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.1.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.1.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.1.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.1.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.1.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.1.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.10.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.10.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.10.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.10.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.10.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.10.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.10.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.10.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.10.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.11.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.11.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.11.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.11.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.11.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.11.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.11.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.11.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.11.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.12.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.12.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.12.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.12.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.12.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.12.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.12.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.12.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.12.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.13.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.13.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.13.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.13.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.13.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.13.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.13.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.13.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.13.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.14.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.14.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.14.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.14.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.14.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.14.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.14.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.14.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.14.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.15.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.15.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.15.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.15.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.15.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.15.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.15.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.15.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.15.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.16.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.16.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.16.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.16.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.16.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.16.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.16.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.16.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.16.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.17.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.17.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.17.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.17.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.17.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.17.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.17.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.17.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.17.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.18.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.18.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.18.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.18.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.18.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.18.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.18.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.18.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.18.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.19.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.19.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.19.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.19.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.19.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.19.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.19.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.19.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.19.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.2.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.2.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.2.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.2.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.2.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.2.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.2.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.2.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.2.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.20.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.20.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.20.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.20.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.20.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.20.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.20.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.20.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.20.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.21.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.21.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.21.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.21.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.21.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.21.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.21.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.21.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.21.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.22.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.22.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.22.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.22.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.22.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.22.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.3.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.3.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.3.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.3.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.3.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.3.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.3.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.3.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.3.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.4.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.4.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.4.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.4.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.4.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.4.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.4.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.4.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.4.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.5.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.5.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.5.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.5.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.5.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.5.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.5.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.5.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.5.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.6.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.6.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.6.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.6.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.6.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.6.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.6.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.6.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.6.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.7.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.7.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.7.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.7.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.7.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.7.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.7.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.7.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.7.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.8.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.8.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.8.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.8.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.8.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.8.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.8.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.8.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.8.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.9.attn.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.9.attn.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.9.attn.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.9.attn.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.9.attn.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.9.attn.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.9.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.9.img_mlp.out.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "transformer_blocks.9.img_mlp.proj.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "txt_in.in_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "txt_in.out_layer.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "txt_in.text_norm.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
+    "norm_out.linear.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "proj_out.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.22.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.22.img_mlp.out.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.22.img_mlp.proj.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.23.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.23.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.23.attn.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.23.attn.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.23.attn.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.23.attn.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.23.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.23.img_mlp.out.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.23.img_mlp.proj.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.24.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.24.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.24.attn.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.24.attn.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.24.attn.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.24.attn.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.24.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.24.img_mlp.out.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.24.img_mlp.proj.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.25.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.25.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.25.attn.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.25.attn.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.25.attn.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.25.attn.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.25.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.25.img_mlp.out.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.25.img_mlp.proj.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.26.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.26.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.26.attn.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.26.attn.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.26.attn.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.26.attn.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.26.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.26.img_mlp.out.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.26.img_mlp.proj.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.27.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.27.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.27.attn.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.27.attn.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.27.attn.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.27.attn.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.27.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.27.img_mlp.out.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.27.img_mlp.proj.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.28.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.28.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.28.attn.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.28.attn.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.28.attn.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.28.attn.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.28.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.28.img_mlp.out.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.28.img_mlp.proj.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.29.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.29.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.29.attn.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.29.attn.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.29.attn.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.29.attn.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.29.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.29.img_mlp.out.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.29.img_mlp.proj.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.30.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.30.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.30.attn.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.30.attn.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.30.attn.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.30.attn.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.30.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.30.img_mlp.out.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.30.img_mlp.proj.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.31.attn.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.31.attn.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.31.attn.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.31.attn.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.31.attn.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.31.attn.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.31.img_mlp.gate_layer.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.31.img_mlp.out.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
+    "transformer_blocks.31.img_mlp.proj.weight": "diffusion_pytorch_model-00002-of-00002.safetensors"
+  }
+}

vae/config.json

@@ -0,0 +1,161 @@
+{
+  "_class_name": "AutoencoderKLQwenImage21",
+  "_diffusers_version": "0.37.0.dev0",
+  "attn_scales": [],
+  "base_dim": 96,
+  "decoder_base_dim": 144,
+  "dim_mult": [
+    1,
+    2,
+    4,
+    8,
+    8
+  ],
+  "dropout": 0.0,
+  "in_channels": 4,
+  "is_residual": true,
+  "latents_mean": [
+    0.5126,
+    0.7721,
+    -0.0631,
+    1.3506,
+    -0.7855,
+    -2.1025,
+    -0.3458,
+    1.3722,
+    1.8873,
+    -1.7177,
+    -0.651,
+    0.2732,
+    0.7562,
+    -0.6163,
+    -1.0277,
+    3.8363,
+    2.021,
+    0.0472,
+    0.932,
+    2.0087,
+    2.4954,
+    -0.1391,
+    -1.4249,
+    1.8464,
+    -0.5236,
+    1.2826,
+    3.7046,
+    -1.3035,
+    2.7286,
+    -1.4518,
+    -1.9036,
+    -1.9955,
+    -0.0342,
+    -1.0265,
+    -0.7636,
+    3.0555,
+    0.0746,
+    -3.0751,
+    -0.1076,
+    1.7376,
+    -1.0914,
+    -1.9435,
+    -0.2784,
+    -1.368,
+    0.4809,
+    -0.4433,
+    0.3764,
+    0.5729,
+    -2.0595,
+    1.096,
+    -1.326,
+    -2.0211,
+    -5.0179,
+    0.5275,
+    4.0162,
+    1.8505,
+    0.3026,
+    1.9373,
+    1.4937,
+    0.2632,
+    0.5547,
+    -1.7121,
+    -0.1562,
+    0.0304
+  ],
+  "latents_std": [
+    3.2001,
+    3.2936,
+    3.4321,
+    3.0091,
+    3.1061,
+    4.0379,
+    4.0705,
+    3.791,
+    3.0785,
+    3.65,
+    3.9308,
+    3.0904,
+    2.8778,
+    3.7675,
+    3.732,
+    5.0756,
+    3.2864,
+    4.0397,
+    3.1317,
+    4.0443,
+    2.9249,
+    3.9454,
+    3.0988,
+    4.2489,
+    3.4896,
+    3.8513,
+    3.9323,
+    3.4719,
+    3.7498,
+    4.283,
+    3.5694,
+    4.2467,
+    3.9037,
+    3.2947,
+    5.077,
+    3.5075,
+    3.27,
+    3.4767,
+    2.8063,
+    5.1125,
+    3.5327,
+    4.7833,
+    3.1286,
+    4.1819,
+    3.8527,
+    3.8312,
+    3.5605,
+    4.3875,
+    3.9624,
+    4.0168,
+    3.5643,
+    4.055,
+    5.5614,
+    4.2963,
+    4.408,
+    3.4959,
+    3.8747,
+    3.7608,
+    3.5735,
+    3.149,
+    3.7662,
+    3.6746,
+    3.4563,
+    3.8161
+  ],
+  "num_res_blocks": 2,
+  "out_channels": 4,
+  "patch_size": null,
+  "scale_factor_spatial": 16,
+  "scale_factor_temporal": 8,
+  "temperal_downsample": [
+    false,
+    true,
+    true,
+    true
+  ],
+  "z_dim": 64
+}

vae/diffusion_pytorch_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a07a1b7c4ee2966a1b3bdc37de9b4f983d56937e46619f709a80b6e490675417
+size 1350989512