| --- |
| license: other |
| license_name: qwen-research |
| license_link: https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE |
| language: |
| - en |
| - zh |
| tags: |
| - comfyui |
| - gguf |
| - qwen |
| - qwen2 |
| - qwen3 |
| - qwen-image |
| - qwen-image-2.1 |
| - text-to-image |
| - image-to-image |
| - image-generation |
| - quantized |
| - diffusion |
| pipeline_tag: text-to-image |
| base_model: Qwen/Qwen-Image-2.1 |
| base_model_relation: quantized |
| --- |
| |
| # Qwen-Image 2.1 GGUF (ComfyUI) |
|
|
| <div align="center"> |
|
|
| [](https://huggingface.co/AlperKTS/Qwen-Image-2.1-GGUF) |
| [](https://huggingface.co/AlperKTS/Qwen-Image-2.1-GGUF) |
| [](https://github.com/city96/ComfyUI-GGUF) |
| [](https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE) |
|
|
| </div> |
|
|
| Quantized GGUF weights of the state-of-the-art **Qwen-Image 2.1** diffusion model for **ComfyUI** using [ComfyUI-GGUF](https://github.com/city96/ComfyUI-GGUF). |
|
|
| These models allow running Qwen-Image 2.1 on consumer GPUs with significantly reduced VRAM footprint while preserving maximum visual quality, text-rendering capabilities, and composition fidelity. |
|
|
| --- |
|
|
| ## 📦 Quantization Versions |
|
|
| | File | Size | VRAM Recommended | Description | |
| | :--- | :--- | :--- | :--- | |
| | **`qwen_image_2.1_Q8_0.gguf`** | ~7.2 GB | 12GB+ | Near-lossless precision. Best quality & fidelity. | |
| | **`qwen_image_2.1_Q6_K.gguf`** | ~5.5 GB | 8GB - 12GB | Optimal balance of high quality and memory efficiency. | |
| | **`qwen_image_2.1_Q5_K_M.gguf`** | ~4.6 GB | 8GB | Fast inference, lowest VRAM consumption, excellent quality. | |
| |
| *Note: The original BF16 diffusion model is ~14.2 GB.* |
| |
| --- |
| |
| ## 🚀 How to Use in ComfyUI |
| |
| ### 1. Install Prerequisites |
| Make sure you have [ComfyUI-GGUF](https://github.com/city96/ComfyUI-GGUF) installed in your `custom_nodes` folder: |
| ```bash |
| cd ComfyUI/custom_nodes |
| git clone https://github.com/city96/ComfyUI-GGUF |
| ``` |
| |
| ### 2. Download Required Files |
| |
| 1. **Diffusion Model (GGUF):** |
| - Download any of the `.gguf` files from this repo (e.g. `qwen_image_2.1_Q8_0.gguf`). |
| - Place it in: `ComfyUI/models/diffusion_models/` (or `ComfyUI/models/unet/`) |
| |
| 2. **VAE:** |
| - Download `qwen_image_2.1_vae_bf16.safetensors` from [Comfy-Org/Qwen-Image-2.1](https://huggingface.co/Comfy-Org/Qwen-Image-2.1/tree/main). |
| - Place it in: `ComfyUI/models/vae/` |
|
|
| 3. **Text Encoder / CLIP:** |
| - Download `qwen3vl_8b_fp8_scaled.safetensors` or `qwen3vl_8b_int8_convrot.safetensors` (or `qwen3vl_8b_bf16.safetensors`) from [Comfy-Org/Qwen-Image-2.1](https://huggingface.co/Comfy-Org/Qwen-Image-2.1/tree/main). |
| - Place it in: `ComfyUI/models/text_encoders/` (or `ComfyUI/models/clip/`) |
| - *Note: In ComfyUI's `CLIPLoader`, select loader type `qwen_image`.* |
|
|
| --- |
|
|
| ## 🎨 Workflows |
|
|
| Pre-configured workflows are included in the [`workflows/`](./workflows) folder: |
| - **`Qwen_Image_2.1_GGUF_Text_to_Image.json`**: Ready-to-run ComfyUI text-to-image workflow using `UnetLoaderGGUF`. |
| - **`Qwen_Image_2.1_GGUF_Image_Edit.json`**: Ready-to-run ComfyUI image-to-image / editing workflow using `UnetLoaderGGUF`. |
| - Drag and drop either workflow JSON directly into ComfyUI to start generating immediately. |
| |
| ### Key Node Configuration |
| - Node: **`UnetLoaderGGUF`** (from ComfyUI-GGUF) |
| - `unet_name`: Select `qwen_image_2.1_Q8_0.gguf` (or `Q6_K` / `Q5_K_M`) |
| - Connect to standard KSampler / ModelSamplingDiscrete nodes. |
| |
| --- |
| |
| ## 📜 Credits & License |
| - Original model created by the **Qwen Team (Alibaba)**. |
| - GGUF conversion & quantization by **AlperKTS**. |
| - Compatible with [ComfyUI-GGUF](https://github.com/city96/ComfyUI-GGUF) by [city96](https://github.com/city96). |
| - Licensed under the original Qwen Research License. |
| |
