Qwen-Image-2512 ComfyUI Native Workflow Usage Guide

Qwen-Image-2512 is the December 2024 update of the Qwen-Image model released by Alibaba’s Tongyi Lab. Compared to the base version released in August, this new version includes multiple improvements in image generation quality.

Key Improvements:

• Enhanced Human Realism: Generated human images look more realistic and natural, with reduced obvious AI-generated artifacts
• Finer Natural Detail: More detailed rendering of landscapes, animal fur, and other natural elements
• Improved Text Rendering: Text in generated images is clearer and more accurate, with better integration between text and images

Related Resources:

• GitHub Project Page
• Hugging Face Model Page
• ModelScope Model Repository

Qwen-Image-2512 ComfyUI Native Workflow Guide

The workflow includes two subgraphs:

• Text to Image (Qwen-Image 2512): Standard 50-step generation process
• Text to Image (Qwen-Image 2512 4steps): 4-step fast generation using Lightning LoRA

1. Workflow File

After updating ComfyUI, you can find the workflow file from the templates, or drag the workflow file below into ComfyUI to load it.

2. Model Download

All model files can be found on Hugging Face or ModelScope.

Text Encoders

• qwen_2.5_vl_7b_fp8_scaled.safetensors

LoRA Model (Optional - for 4-step acceleration)

If you want faster generation speed, you can use Lightning LoRA to reduce the generation steps from 50 to 4:

• Qwen-Image-Lightning-4steps-V1.0.safetensors

Diffusion Models

You can choose one of the following two models based on your needs:

• qwen_image_2512_fp8_e4m3fn.safetensors (Recommended for most users)
• qwen_image_2512_bf16.safetensors (If you have sufficient VRAM and want better image quality, you can choose this version)

VAE Model

• qwen_image_vae.safetensors

3. Model File Storage Location

After downloading, please place the model files according to the following directory structure:

📂 ComfyUI/
├── 📂 models/
│   ├── 📂 text_encoders/
│   │   └── qwen_2.5_vl_7b_fp8_scaled.safetensors
│   ├── 📂 loras/
│   │   └── Qwen-Image-Lightning-4steps-V1.0.safetensors  (Optional)
│   ├── 📂 diffusion_models/
│   │   ├── qwen_image_2512_fp8_e4m3fn.safetensors        (or qwen_image_2512_bf16.safetensors)
│   └── 📂 vae/
│       └── qwen_image_vae.safetensors

4. Supported Image Aspect Ratios

Qwen-Image-2512 supports multiple image aspect ratios. You can choose the appropriate size based on your needs: