Skip to content

qwen-image-edit

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

qwen-image-edit

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
.gitignore
.gitignore
 
 
README.md
README.md
 
 
gpu.jsonc
gpu.jsonc
 
 
install_comfyui.py
install_comfyui.py
 
 
run.py
run.py
 
 
sample-image.png
sample-image.png
 
 
startup.sh
startup.sh
 
 

README.md

ComfyUI with Qwen Image Edit

GPU CLI template for running ComfyUI with workflow preloading. Specify any ComfyUI workflow URL and the setup script will download it along with all required models.

Quick Start

# Launch ComfyUI (setup runs automatically on first start)
gpu use .

Open http://localhost:8188 to access ComfyUI once the service is ready.

Hardware Requirements

  • GPU: RTX 4090 or equivalent (24GB VRAM minimum)
  • Storage: ~30GB for models + workspace
  • Default config: gpu_type: "NVIDIA GeForce RTX 4090"

Using a Different Workflow

Edit gpu.jsonc and change the workflow URL:

{
  "workflow": "https://raw.githubusercontent.com/Comfy-Org/workflow_templates/refs/heads/main/templates/image_qwen_image_edit.json"
}

The setup script will:

  1. Download your workflow
  2. Parse it to find required models (UNETLoader, CLIPLoader, VAELoader, LoraLoader nodes)
  3. Download all models from HuggingFace
  4. Place the workflow in ComfyUI/user/default/workflows/

Finding Workflows

Default Workflow: Qwen Image Edit

The default workflow uses Qwen-Image-Edit, a 20B parameter model for precise image editing with natural language.

Capabilities:

  • Edit images with text prompts ("add sunglasses", "change sky to sunset")
  • Precise text editing within images
  • Style transfer and appearance changes
  • Object addition/removal

Example prompts:

  • "make the sky overcast and add rain"
  • "add sunglasses to the person"
  • "change the car color to red"
  • "remove the background"

Models downloaded (~26GB):

Model Size Purpose
qwen_image_edit_fp8_e4m3fn.safetensors ~20GB Main diffusion model
qwen_2.5_vl_7b_fp8_scaled.safetensors ~5GB Text encoder
qwen_image_vae.safetensors ~1GB VAE
Qwen-Image-Edit-Lightning-4steps-V1.0-bf16.safetensors ~500MB LoRA for fast 4-step generation

ComfyUI Keyboard Shortcuts

Key Action
Ctrl+Enter Queue prompt
Ctrl+Shift+Enter Queue prompt (front)
Ctrl+Z Undo
Ctrl+Y Redo
Space Pan canvas
Ctrl+M Mute/unmute node
Ctrl+B Bypass node

Troubleshooting

"Model not found" error

Re-run the template to re-download models:

gpu use .

Out of memory (OOM)

The Qwen models require ~24GB VRAM. Ensure you're using an RTX 4090 or equivalent:

{
  "gpu_type": "NVIDIA GeForce RTX 4090",
  "min_vram": 24
}

Port already in use

ComfyUI defaults to port 8188. The GPU CLI will remap it automatically.

Workflow not loading

Check that your workflow URL returns valid JSON. Test by opening the URL in a browser.

Files

  • gpu.jsonc - GPU CLI configuration with workflow URL
  • startup.sh - Startup script (runs setup + starts ComfyUI)
  • install_comfyui.py - Downloads ComfyUI, models, and workflow
  • run.py - Starts the ComfyUI server (for manual use)
  • sample-image.png - Test image for editing workflows

Output Syncing

Generated images are saved to ComfyUI/output/ and automatically synced back to your local machine.