Switch to conradlocke/krea2-identity-edit instruction edit LoRA

README.md

@@ -1,49 +1,42 @@
 ---
 title: Krea 2 Identity Edit
-emoji: 🧬
-colorFrom: blue
-colorTo: gray
+emoji: ✏️
+colorFrom: pink
+colorTo: green
 sdk: gradio
 sdk_version: 6.20.0
 app_file: app.py
 pinned: false
 hardware: zero-a10g
+short_description: Instruction-based identity-preserving photo editing
 python_version: "3.10"
-startup_duration_timeout: 45m
-short_description: Identity-preserving instruction image editing on Krea 2
+startup_duration_timeout: "30m"
 models:
   - krea/Krea-2-Turbo
+  - ostris/Krea2OstrisEdit
   - conradlocke/krea2-identity-edit
 ---
 
-# 🧬 Krea 2 Identity Edit
+# Krea 2 Identity Edit ✏️
 
-Instruction-based, **identity-preserving** image editing built on
-[Krea 2 Turbo](https://huggingface.co/krea/Krea-2-Turbo) with the community LoRA
-[`conradlocke/krea2-identity-edit`](https://huggingface.co/conradlocke/krea2-identity-edit).
-
-Give it an image and a plain-language instruction; it edits while preserving what you
-didn't ask to change — including the person's likeness. Person re-staging, local edits
-(recolor / add / remove / replace), replace-with-reference, and full-image restyles.
+Instruction-based, **identity-preserving** image editing on [Krea 2 Turbo](https://huggingface.co/krea/Krea-2-Turbo),
+using the [Krea2 Identity Edit LoRA](https://huggingface.co/conradlocke/krea2-identity-edit) (v1.1)
+served through the [Krea2OstrisEdit](https://huggingface.co/ostris/Krea2OstrisEdit) community pipeline.
 
 ## How it works
 
-Krea 2 is a text-to-image MMDiT; this LoRA turns it into an instruction editor via a
-**dual-conditioning** recipe that stock `Krea2Pipeline` does not provide. This Space
-reproduces the two custom pieces from the reference
-[ComfyUI-Krea2Edit](https://github.com/lbouaraba/comfyui-krea2edit) node pack in plain
-diffusers / PyTorch:
+Upload a photo and give a plain-language instruction. The pipeline conditions on your image via
+dual conditioning — in-context VAE tokens + image-grounded Qwen3-VL encoding — and the identity-edit
+LoRA is trained to consume it, editing what you asked while keeping the face, outfit, and untouched
+scene elements fixed.
 
-- **Grounded encode** — the instruction is encoded *together with the source image*
-  through the Qwen3-VL text encoder (image tokens inserted via the grounded chat
-  template), tapping the 12 selected decoder layers.
-- **Source patch** — the VAE-encoded source latent is prepended to the transformer
-  sequence as clean tokens (RoPE frame index: source = 1, target = 0), so appearance is
-  carried in-context and only the target tokens are denoised.
+- **Re-stage a person:** "create a photo of this person at a night market" (same face + outfit, relit)
+- **Local edits:** add / remove / replace objects, recolor, outfit changes
+- **Restyle:** global style with preserved composition
 
-Default recipe: Turbo, 8 steps, guidance off (Krea convention). `grounding_px` trades
-edit adherence (lower) against identity/likeness (higher); 768 is balanced, 1024+ for
-people. For removals / large deletions, raise steps and guidance.
+## Notes
 
-*An unofficial community fine-tune of Krea 2 Raw — not affiliated with or endorsed by
-Krea.ai, Inc. Weights are distributed under the Krea 2 Community License.*
+- The output aspect ratio auto-matches your source photo (training pairs are same-size), targeting ~1MP.
+- Recommended: 8–12 steps, CFG 1.0 for most edits; ~20 steps / CFG 3.0 for large removals.
+- Single LoRA replaces the pipeline's default style-reference LoRA (both are the edit brain); other
+  style/character LoRAs can stack on top.