Have you ever tried making a small edit to an image and ended up reworking half the picture? You remove an unwanted object, and suddenly the background looks glitchy.
Or may be you just want to change a word in an image without touching the font, size, or style but the editing tool either can't handle it or makes it look totally off. Qwen Image Edit released by Alibaba built on the powerful 20B Qwen Image model, leverages a dual system approach that supports both English and Chinese.
 |
| Ref- Official Qwen image Edit page |
It feeds images into Qwen2.5 VL for semantic understanding (so the model knows what's happening in the image) and into a VAE Encoder for visual appearance control (so it knows how the image looks).
 |
| Ref- Official Qwen Image edit page |
In simpler terms, it can balance both meaning and style, which is exactly why it outperforms existing tools in tricky edits. You can access their research paper for more detailed findings.
Installation
1. Get ComfyUI into your system. Update it if you already installed by navigating to the Manager tab then select Update All.
2. Download Qwen Image Edit FP8 (qwen_image_edit_fp8_e4m3fn.safetensors) and save it into your ComfyUI/models/diffusion_models directory.
3. Download Text encoder (qwen_2.5_vl_7b_fp8_scaled.safetensors) and save it into your ComfyUI/models/text_encoders directory.
4. Download Vae (qwen_image_vae.safetensors) and save it into ComfyUI/models/vae folder.
5. Download Lora model (Qwen-Image-Lightning-4steps-V1.0.safetensors) if you want then save it into your ComfyUI/models/loras directory.
6. Restart ComfyUI and refresh to take effect.
Qwen Image Edit GGUF
If having low VRAMs use the GGUF variant.
1. Update the Custom nodes ComfyUI-GGUF by City 96. If not installed yet, then install it from the Manager>Custom nodes manager, then search for ComfyUI-GGUF by City96.
2. Download the Qwen Image Edit GGUF Model (by Quanstack) from Hugging face repository and save it into your ComfyUI/models/unet folder.
3. Download Text encoder (qwen_2.5_vl_7b_fp8_scaled.safetensors) and save it into your ComfyUI/models/text_encoders directory.
4. Finally, download Vae (qwen_image_vae.safetensors) and put it into ComfyUI/models/vae folder.
5. Download Lora model (Qwen-Image-Lightning-4steps-V1.0.safetensors) if you want then save it into your ComfyUI/models/loras directory.
6. Restart ComfyUI and refresh to take effect.
Workflow
1. Download the workflow (Qwen_Image_Edit.json ) from our Hugging Face repository.
You can also get the workflow from Comfyui dashboard. To open it just select on Workflow>Browse Templates>Image>Qwen Image Edit workflow.
2. Drag and drop into ComfyUI.
(a) Load your target image to edit into Load image node.
(b) Load Qwen Image edit model into Load diffusion model node or Unet loader if using GGUF model variant.
(c) Load Clip model and VAE.
(d) Select Lora model.
(e) Put the relevant prompt to edit into the prompt box.
(f) Set the KSampler Settings.
Basic model FP8
Step-20 (Official -50)
CFG-4.0
Sampler -Euler
fp8 + Lightning LoRA:
Steps-4
CFG-1.0
Sampler -Euler
(g) Hit Run to execute the workflow.
