--- license: other license_name: qwen-research base_model: - abenzerps/Qwen-Image-2.1-Uncensored-GGUF base_model_relation: quantized pipeline_tag: text-to-image library_name: gguf tags: - gguf - qwen - image-generation - comfyui - comfyui-gguf - genesis --- # Qwen-Image-2.1-Uncensored-GGUF -> Genesis > ⚡ If you like this Genesis LLM release you can donate to me via [Hipolink](https://hipolink.net/luffythefox) or: > USDT (TRC20): `TGa4KTwHfF6zDBsLUBEjd1f1KdeAFwYUks` > USDT (ERC20): `0x93F4019E0aa85d8078F56B3D3176Fab5Dfa79924` > USDT (SOL): `BorkkyPDG4aRN2op8c5SQF5NithX38U4sh7wDAWQhYyX` > and support future Genesis LLM development. > ⚡ **Why Genesis project exists?** During training, **ALL** models don't just learn knowledge - they also accumulate random noise in their tensors. This noise builds up and creates something I call the **Noise Gate** - a fundamental barrier that stops LLM models from learning further and makes them unstable, verbose, and prone to hallucinations. My approach reduces this noise. It repairs the signal in tensors without touching the learned knowledge and gradient using [Marchenko–Pastur distribution](https://en.wikipedia.org/wiki/Marchenko%E2%80%93Pastur_distribution) as a core criteria. The result is a model that consistent in performance, context clarity and following instructions, because it's no longer fighting its own internal chaos. > **What is Genesis?** Genesis is post training data regeneration and calibrarion algorythm for neural networks (LLM) in GGUF format that I made with AI help during almost half a year of development. It's optimized, architecture independent, works with any model in GGUF format and based on mathematical statistics. I don't train or finetune models, I repair **purity of signal** in them instead on Google Collab Free on Tesla T4 GPU via Python based on how models learns information. On first stage I scan ssm_conv1d tensors in model, they handle long context memory. I repair balance between heads in them. On second stage, I scan blocks in model via chunks via 3 parameters and pick best one that fits to weight distribution in tensor. Best picked chunk replaces zero chunks in broken tensor without touching learned structure in model. On third stage I scan model and detect noise in tensors via custom SVD. During scanning I exclude token_embd.weight, output.weight, 1D tensors, bias and norms. Then I reduce training noise in tensors via custom SVD based on [Marchenko–Pastur law](https://en.wikipedia.org/wiki/Marchenko%E2%80%93Pastur_distribution) with preserved training data, 99% of siginal and learned gradient. `Many tensors in this model in original weights shared by Alibaba were singular with huge condition number for matrices. They distorted the signal distribution between tensors instead of transmitting it correctly. I fixed it as much as I can for base model and text encoder and reduced condition number for matrices for stable inference during image generation.` ## Any questions? > Contact: `luffythefox@mail.ru`, `azakharchenko92@gmail.com` > My Telegram: `@LuffyTheFox` GGUF quantizations of [abenzerps/Qwen-Image-2.1-Uncensored-GGUF](https://huggingface.co/abenzerps/Qwen-Image-2.1-Uncensored-GGUF) for local image generation using the modified via Genesis base weights. ## Usage Use the model with [ComfyUI](https://github.com/comfyanonymous/ComfyUI) and [ComfyUI-GGUF](https://github.com/leejet/ComfyUI-GGUF). All required companion files (GGUF transformer, text encoder, and VAE) are hosted directly in this repository. ### 1. Download & File Placement Download the files and place them in their respective ComfyUI directories: ```text ComfyUI/ └── models/ ├── diffusion_models/ │ └── qwen-image-2.1-UC-Q4_K_M.gguf # Choose one GGUF quantization (Q4_K_M recommended) ├── text_encoders/ │ └── qwen3vl_8b_bf16.safetensors # Or qwen3vl_8b_int8_convrot.safetensors (recommended for lower memory) └── vae/ └── qwen_image_2.1_vae_bf16.safetensors ``` ### 2. ComfyUI Setup 1. **Install ComfyUI-GGUF**: Use the maintained fork with native Qwen-Image 2.1 support by cloning [leejet/ComfyUI-GGUF](https://github.com/leejet/ComfyUI-GGUF) into your custom nodes: ```bash cd ComfyUI/custom_nodes git clone https://github.com/leejet/ComfyUI-GGUF ``` *(Note: If you have the older `city96/ComfyUI-GGUF` installed and encounter an `Unknown model architecture!` error, update to the `leejet` fork above or add `ModelQwenImage` to `tools/convert.py`).* 2. **Node Configuration**: - **Diffusion Model**: Add the **`Unet Loader (GGUF)`** node and select your downloaded `.gguf` file. - **Text Encoder**: Add the standard **`CLIPLoader`** node, select `qwen3vl_8b_bf16.safetensors` (or `int8`), and set **`type`** to **`qwen_image`**. - **VAE**: Add the standard **`VAELoader`** node and select `qwen_image_2.1_vae_bf16.safetensors`. 3. **Official Workflows**: - You can use the official Comfy-Org workflow templates: [Text-to-Image](https://github.com/Comfy-Org/workflow_templates/blob/main/templates/image_qwen_image_2_1_t2i.json) or [Image Edit](https://github.com/Comfy-Org/workflow_templates/blob/main/templates/image_qwen_image_2_1_image_edit.json). - In the workflow, simply replace the default `UNETLoader` node with **`Unet Loader (GGUF)`**. ### Memory & Performance Notes - **Optimal Setup (GPU + RAM)**: Keep the **GGUF diffusion model in GPU VRAM** (where speed is crucial during sampling) and let the **text encoder run in / offload to System RAM (CPU)**. Because text encoding only runs once per prompt, this saves 9–17 GB of VRAM with virtually zero impact on generation speed. - **Recommended Configuration**: - **Diffusion**: `NVFP4` - **Text Encoder**: `NVFP4` - **Low VRAM Mode**: If you experience VRAM out-of-memory errors, start ComfyUI with the `--lowvram` argument. ## Source and build - **Source model:** [Qwen/Qwen-Image-2.1](https://huggingface.co/Qwen/Qwen-Image-2.1) - **Text encoder & VAE source:** [Comfy-Org/Qwen-Image-2.1](https://huggingface.co/Comfy-Org/Qwen-Image-2.1) - **Source revision:** `b3179ad355be050328e483a9dfdd9e60cd62adfa` - **Conversion:** [stable-diffusion.cpp](https://github.com/leejet/stable-diffusion.cpp) commit `1330cebae8f2ba99249df846cc0c9444fcbd4308` - **License:** Qwen Research License - **Checksums:** [SHA256SUMS](./SHA256SUMS)