AI Image Editing with Qwen-Image-Edit-2511¶

The Image Editor integrates Qwen-Image-Edit-2511, a powerful AI diffusion model for text-guided image editing.

Overview¶

Qwen-Image-Edit-2511 enables natural language image editing without masks or manual selection. Simply describe what you want to change, and the AI handles the rest.

Key Capabilities: - Object addition, removal, and replacement - Background replacement and removal - Style transfer (Ghibli, Pixar, anime, etc.) - Text editing on images - Clothing and accessory modifications - Season and lighting changes - Photo restoration and enhancement

Using AI Edit in the App¶

When deployed with enable_qwen_image_edit = true, the Image Editor includes an AI Edit panel in the sidebar.

Quick Start¶

Load or create an image - Use "Load Image" to import an image, or draw/create content on the canvas
Enter a prompt - In the AI Edit panel, describe what you want to change
Click "Apply AI Edit" - Wait 10-30 seconds for processing
View the result - The edited image replaces the canvas content

AI Edit Panel Features¶

Prompt Input - Enter natural language instructions like "remove the background" or "add a sunset sky" - Keep prompts concise and action-oriented

Quick Suggestions - Click suggestion chips for common edits: - Remove the background - Add a sunset sky background - Transform into Studio Ghibli style - Make it look like a watercolor painting

Advanced Options (click to expand) - Negative prompt: Describe what to avoid (e.g., "blurry, low quality") - Quality steps: Higher values (40-50) = better quality but slower - Seed: Set a number for reproducible results

Environment Configuration¶

You can configure the API URL via environment variable:

# In .env or environment
VITE_DIFFUSION_API_URL=http://your-gpu-server:8000

If not set, the app defaults to http://<current-host>:8000.

API Endpoints¶

When enable_qwen_image_edit = true, the following API endpoints are available:

Health Check¶

GET /health

Returns server status and model loading state.

Model Info¶

GET /info

Returns model variant, GPU info, and memory usage.

Edit Image (Multipart Form)¶

POST /edit
Content-Type: multipart/form-data

Parameters:
- image: Image file (PNG, JPG)
- prompt: Edit instruction (required)
- negative_prompt: What to avoid (optional)
- num_inference_steps: Quality steps (default: 40)
- guidance_scale: Prompt adherence (default: 1.0)
- true_cfg_scale: Consistency control (default: 4.0)
- seed: For reproducibility (optional)

Edit Image (Base64)¶

POST /edit/base64
Content-Type: application/json

{
  "image": "<base64-encoded-image>",
  "prompt": "your edit instruction",
  "negative_prompt": null,
  "num_inference_steps": 40,
  "guidance_scale": 1.0,
  "true_cfg_scale": 4.0,
  "seed": null
}

Prompt Examples¶

Object Removal¶

"remove the person on the right"
"remove the trash can from the scene"
"remove the watermark"

Object Addition¶

"add a red umbrella in her hand"
"add a cat sitting on the couch"
"add sunglasses to the person"

Background Changes¶

"replace background with studio white"
"change the background to a sunset beach"
"replace background with a futuristic cityscape at night"

Style Transfer¶

"transform into Studio Ghibli anime style"
"convert to Pixar 3D animation style"
"make it look like a watercolor painting"
"transform into pixel art style"

Clothing & Accessories¶

"change the t-shirt to a formal suit"
"add a cowboy hat"
"change the dress color to red"

Lighting & Environment¶

"use soft light to relight the image"
"change from day to night"
"convert summer scene to winter with snow"
"add dramatic sunset lighting"

Text Editing¶

"change the sign text to 'Welcome'"
"replace the billboard text with 'SALE 50% OFF'"

Photo Enhancement¶

"enhance the image quality"
"remove blur and sharpen details"
"fix the lighting and color balance"

Best Practices¶

Writing Effective Prompts¶

Be Direct: Use action verbs like "replace," "add," "remove," "change"
Good: "remove the person in the background"
Bad: "I would like the background person to not be there"
Be Specific: Name objects and regions explicitly
Good: "change the blue car to red"
Bad: "change the color"
Keep It Concise: 50-200 characters works best
Good: "replace background with a sunny beach, blue sky"
Bad: "I want you to take the current background and replace it with something that looks like a beautiful sunny beach with crystal clear water and a bright blue sky with some fluffy white clouds"
One Edit at a Time: For complex changes, do multiple passes
First: "remove the person on the left"
Then: "add a potted plant where the person was"

Parameter Tuning¶

Parameter	Default	Description
`num_inference_steps`	40	Higher = better quality, slower. Range: 20-50
`true_cfg_scale`	4.0	Controls consistency. Keep at 4.0 for best results
`guidance_scale`	1.0	Prompt adherence. 1.0 works well for edits
`seed`	random	Set for reproducible results

Quality Tips¶

For subtle changes: Use shorter, precise prompts
For creative restyling: Allow broader descriptions
For text on images: The model handles Chinese and English text well
For people: Works best with clear, well-lit subjects

Usage Examples¶

cURL Examples¶

Remove an object:

curl -X POST "http://<ip>:8000/edit" \
  -F "image=@photo.jpg" \
  -F "prompt=remove the person in the background" \
  -o edited.png

Style transfer:

curl -X POST "http://<ip>:8000/edit" \
  -F "image=@photo.jpg" \
  -F "prompt=transform into Studio Ghibli anime style" \
  -F "num_inference_steps=50" \
  -o ghibli_style.png

Background replacement:

curl -X POST "http://<ip>:8000/edit" \
  -F "image=@portrait.jpg" \
  -F "prompt=replace background with a professional studio gradient" \
  -o professional_headshot.png

JavaScript/Fetch Example¶

async function editImage(imageFile, prompt) {
  const formData = new FormData();
  formData.append('image', imageFile);
  formData.append('prompt', prompt);
  formData.append('num_inference_steps', 40);

  const response = await fetch('http://<ip>:8000/edit', {
    method: 'POST',
    body: formData
  });

  if (response.ok) {
    const blob = await response.blob();
    return URL.createObjectURL(blob);
  }
  throw new Error('Edit failed');
}

// Usage
const editedImageUrl = await editImage(file, 'add sunglasses');

Python Example¶

import requests

def edit_image(image_path, prompt, output_path):
    with open(image_path, 'rb') as f:
        response = requests.post(
            'http://<ip>:8000/edit',
            files={'image': f},
            data={
                'prompt': prompt,
                'num_inference_steps': 40,
                'true_cfg_scale': 4.0
            }
        )

    if response.status_code == 200:
        with open(output_path, 'wb') as out:
            out.write(response.content)
        return output_path
    raise Exception(f"Error: {response.text}")

# Usage
edit_image('input.jpg', 'change to winter scene with snow', 'winter.png')