Qwen-Image 4-Step Text/Image-to-Image

Requirements

• GPU: NVIDIA RTX 30XX/40XX/50XX (FP16)
• VRAM: 6 GB minimum
• OS: Windows
• Storage: 40 GB+ free

Model downloads

• qwen-image Q3_K_S GGUF (→ unet) https://huggingface.co/city96/Qwen-Image-gguf/tree/main ⧉
• Qwen2.5-VL-7B-Instruct Q4_0 GGUF (→ clip) https://huggingface.co/unsloth/Qwen2.5-VL-7B-Instruct-GGUF/tree/main ⧉
• qwen_image_vae.safetensors (→ vae) https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/tree/main/split_files/vae ⧉
• Qwen-Image-Lightning 4-steps LoRA https://huggingface.co/lightx2v/Qwen-Image-Lightning/tree/main ⧉
• 2xLexicaRRDBNet_Sharp (upscaler) https://huggingface.co/Thelocallab/2xLexicaRRDBNet_Sharp/tree/main ⧉
• Qwen Image 20B diffusion models https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/tree/main/split_files/diffusion_models ⧉

Custom nodes

• ComfyUI Manager https://github.com/ltdrdata/ComfyUI-Manager ⧉
• ComfyUI_LayerStyle https://github.com/chflame163/ComfyUI_LayerStyle ⧉
• ComfyUI-KJNodes https://github.com/kijai/ComfyUI-KJNodes ⧉
• ComfyUI-GGUF https://github.com/city96/ComfyUI-GGUF ⧉
• ComfyUI-Florence2 https://github.com/kijai/ComfyUI-Florence2 ⧉

Setup steps

1. Install the ComfyUI portable package and the custom nodes above.
2. Download all models to their folders (unet, clip, vae, loras, upscale_models, diffusion_models).
3. Load the Qwen-Image T2I/I2I workflow into ComfyUI.
4. Use a text prompt for text-to-image, or load an image for image-to-image.
5. For inpainting: right-click the image → "open mask editor", mask the area, add a prompt.

Notes & tips

• Inpainting is experimental — adjust denoise in the Basic Scheduler for consistency.
• RunPod template: https://get.runpod.io/Qwen-Image-Template