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 BF16/FP8 (qwen_image_edit_fp8_e4m3fn.safetensors or qwen_image_edit_bf16.safetensors ) and save it into your ComfyUI/models/diffusion_models directory. These will consume high VRAM (at least 24GB). If you have lower VRAM, use the GGUF variant explained below.
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.
 |
| Qwen Image edit Workflow (Click to zoom) |
(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.
Lets try to do some image edits with this model. we will give basic prompting and test how this performs.
Text Rendering
This is our inputted image generated using Qwen Image model.
Prompt: Change the text to 'Stable Diffusion Café'
CFG-4
Steps-40
Sampler-Euler
This is the first attempt we got this as a result. You can see how precise it was when doing editing with text as the letter E having apostrophe also gets processed. So,this has performed better.
Product photography
This is the image of a diet coke and we want to take it as sample input to do some product photography for ads.
Prompt: Energy Drink Can bursting through a wall of ice with an explosion of coke splash behind it, surrounded by energy lightning bolts, cinematic lighting, power and energy theme, commercial photography setup.
CFG-4
Steps-40
Sampler-Euler
Here, everything looks good but one thing you can observe is the color of can has been changed from silver to red. This is not as accepted.
Well, sometimes the model generate wrong results. So, you can also try Qwen Image Edit 2509 Plus that is the more improved version and also follow the prompts better.
