Abiray/Qwen-Image-2.1-GGUF

Qwen-Image-2.1 GGUF Quants

This repository provides quantized GGUF checkpoints for Qwen/Qwen-Image-2.1, optimized for low-VRAM inference in ComfyUI using the ComfyUI-GGUF custom node.

Quantizing the diffusion transformer drastically reduces memory pressure during generation while preserving sharp detail, composition, and prompt alignment.

---

📦 Quantization Breakdown & File Details

Filename	Quant Type	Size	Recommended VRAM / Profile
qwen_image_2.1_Q8_0.gguf	Q8_0	7.59 GB	12 GB+ (Near-lossless fidelity)
qwen_image_2.1_Q6_K.gguf	Q6_K	5.88 GB	10–12 GB (High quality sweet spot)
qwen_image_2.1_Q5_K_M.gguf	Q5_K_M	5.01 GB	8–10 GB (Balanced performance & memory)
qwen_image_2.1_Q4_K_M.gguf	Q4_K_M	4.19 GB	6–8 GB (Standard consumer GPU baseline)
qwen_image_2.1_Q4_K_S.gguf	Q4_K_S	4.06 GB	6–8 GB (Compact 4-bit)
qwen_image_2.1_Q3_K_M.gguf	Q3_K_M	3.19 GB	4–6 GB (Extreme budget / minimum VRAM)

---

🖼️ Sample Generations (Q4_K_M)

Prompt

-

Prompt

-

Prompt

-

Prompt

-

---

🧩 Required Components (Text Encoders & VAE)

To run these models in ComfyUI, you will need the matching text encoders and VAE:

1. Text Encoders:

   • Download from: Comfy-Org/Qwen-Image-2.1 Text Encoders
   • Place files into: ComfyUI/models/clip/ (or ComfyUI/models/text_encoders/)

2. VAE:

   • Download from: Comfy-Org/Qwen-Image-2.1 VAE
   • Place files into: ComfyUI/models/vae/

3. Diffusion Model (This Repo):

   • Download your preferred .gguf file from above.
   • Place into: ComfyUI/models/diffusion_models/ (or ComfyUI/models/unet/)

---

🎨 Included Ready-to-Use Workflows

Both Text-to-Image and Image-to-Image editing workflows are packaged in this repo:

• Text-to-Image: Qwen_Image_2.1_GGUF_Text2Image.json
• Image-to-Image / Editing: Qwen_Image_2.1_GGUF_Image2Image_Edit.json

How to Use:

1. Ensure you have installed ComfyUI-GGUF (search for ComfyUI-GGUF inside the ComfyUI Manager).
2. Drag and drop either of the .json workflow files into your ComfyUI workspace.
3. In the Unet Loader (GGUF) node, select the downloaded .gguf file.
4. Verify your CLIP/Text Encoder and VAE node loaders point to the files downloaded above.
5. Queue prompt and generate!

---

👏 Acknowledgments & Credits

• Base model developed by the Qwen Team / Alibaba Cloud: Qwen/Qwen-Image-2.1.
• ComfyUI integration assets and pipeline weights provided by Comfy-Org.
• GGUF quantization powered by llama.cpp and ComfyUI-GGUF.