Image Generation

Overview

Open WebUI supports powerful image generation and editing capabilities through multiple engines, allowing you to create custom visuals directly in your chat conversations.

Supported Engines

OpenAI
Gemini
ComfyUI
AUTOMATIC1111

DALL·E Integration

DALL·E 2: Fast, cost-effective
DALL·E 3: Higher quality, more detailed
GPT-IMAGE 1 & 1.5: Latest models

{
  "IMAGE_GENERATION_ENGINE": "openai",
  "IMAGE_GENERATION_MODEL": "dall-e-3",
  "IMAGES_OPENAI_API_BASE_URL": "https://api.openai.com/v1",
  "IMAGES_OPENAI_API_KEY": "sk-..."
}

Google Imagen

Imagen 3.0: High-quality generation
Multiple endpoint methods
Batch generation support

{
  "IMAGE_GENERATION_ENGINE": "gemini",
  "IMAGE_GENERATION_MODEL": "imagen-3.0-generate-002",
  "IMAGES_GEMINI_API_BASE_URL": "https://us-central1-aiplatform.googleapis.com/v1",
  "IMAGES_GEMINI_API_KEY": "your-api-key",
  "IMAGES_GEMINI_ENDPOINT_METHOD": "generateContent"
}

Local Workflow Execution

Custom workflows
Full control over generation
Advanced LoRA/embedding support
Local/self-hosted

{
  "IMAGE_GENERATION_ENGINE": "comfyui",
  "COMFYUI_BASE_URL": "http://comfyui:8188",
  "COMFYUI_API_KEY": "optional-key",
  "COMFYUI_WORKFLOW": "{...}",  // JSON workflow
  "COMFYUI_WORKFLOW_NODES": [...]
}

Stable Diffusion WebUI

Popular self-hosted option
Extensive model library
Community extensions
Free and open source

{
  "IMAGE_GENERATION_ENGINE": "automatic1111",
  "AUTOMATIC1111_BASE_URL": "http://stable-diffusion:7860",
  "AUTOMATIC1111_API_AUTH": "username:password",
  "AUTOMATIC1111_PARAMS": {...}
}

Basic Generation

Create images from text descriptions:

Enable Image Generation

{
  "ENABLE_IMAGE_GENERATION": true,
  "IMAGE_GENERATION_ENGINE": "openai",
  "IMAGE_GENERATION_MODEL": "dall-e-3"
}

Configure Parameters

{
  "IMAGE_SIZE": "1024x1024",  // or "auto" for compatible models
  "IMAGE_STEPS": 50,          // Quality/speed tradeoff
  "ENABLE_IMAGE_PROMPT_GENERATION": true
}

Generate in Chat

Simply describe what you want:

Generate an image of a futuristic city at sunset

API Usage

Generate images programmatically:

# POST /api/v1/images/generations
curl -X POST "https://your-instance/api/v1/images/generations" \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A serene mountain landscape at dawn",
    "n": 1,
    "size": "1024x1024",
    "steps": 50,
    "negative_prompt": "blurry, low quality"
  }'

Response:

[
  {
    "url": "/api/v1/files/{file-id}/content"
  }
]

Generation Parameters

Prompt

Text description of desired image

Be specific and detailed
Include style, mood, lighting
Mention artistic techniques

Size

Image dimensions

Standard: 512x512, 1024x1024
Custom: WIDTHxHEIGHT format
“auto”: Model-dependent sizing

Steps

Generation iterations

More steps = higher quality
More steps = slower generation
Typical range: 20-100

Negative Prompt

What to avoid in the image

“blurry, distorted”
“low quality, watermark”
Engine-dependent support

Image Editing

Modify existing images with AI.

Configuration

{
  "ENABLE_IMAGE_EDIT": true,
  "IMAGE_EDIT_ENGINE": "openai",  // openai, gemini, comfyui
  "IMAGE_EDIT_MODEL": "dall-e-2",
  "IMAGE_EDIT_SIZE": "1024x1024"
}

Edit Workflow

Upload Image

Provide source image(s):

File upload
URL reference
File ID from previous generation

Describe Changes

Specify desired modifications:

{
  "prompt": "Add a rainbow in the sky",
  "image": "data:image/png;base64,..."
}

Optional Parameters

Background color for transparency
Multiple image inputs (for Gemini)
Size adjustments

Generate Edited Version

API returns modified image URL

Edit API

# POST /api/v1/images/edit
curl -X POST "https://your-instance/api/v1/images/edit" \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "image": "https://example.com/image.png",
    "prompt": "Change the sky to nighttime with stars",
    "size": "1024x1024",
    "n": 1
  }'

Supported Edit Operations

OpenAI
Gemini
ComfyUI

DALL·E Edit Capabilities:

Single image input
Inpainting/outpainting
Background replacement
Style transfer
Detail enhancement

ComfyUI Integration

Advanced users can leverage ComfyUI’s powerful workflow system.

Workflow Configuration

Design Workflow

Create workflow in ComfyUI interface:

Load desired models
Configure nodes
Test generation

Export Workflow

Save workflow as JSON from ComfyUI

Configure Nodes

Map workflow nodes to Open WebUI parameters:

{
  "COMFYUI_WORKFLOW_NODES": [
    {
      "type": "model",
      "node_ids": ["4"]
    },
    {
      "type": "prompt",
      "node_ids": ["6", "7"]
    },
    {
      "type": "size",
      "node_ids": ["5"]
    }
  ]
}

Upload to Open WebUI

Paste workflow JSON in settings:

{
  "COMFYUI_WORKFLOW": "{...}"
}

Workflow Node Types

model: Checkpoint/model selector
prompt: Positive/negative prompts
size: Width/height dimensions
steps: Sampling steps
seed: Random seed control
sampler: Sampling method

ComfyUI workflows allow complete control over the generation pipeline, including custom samplers, LoRAs, and post-processing.

AUTOMATIC1111 Integration

Configuration

{
  "IMAGE_GENERATION_ENGINE": "automatic1111",
  "AUTOMATIC1111_BASE_URL": "http://localhost:7860",
  "AUTOMATIC1111_API_AUTH": "username:password",  // Optional
  "IMAGE_SIZE": "512x512",
  "IMAGE_STEPS": 50
}

Custom Parameters

Pass additional parameters to AUTOMATIC1111:

{
  "AUTOMATIC1111_PARAMS": {
    "cfg_scale": 7.5,
    "sampler_name": "Euler a",
    "enable_hr": true,
    "hr_scale": 2,
    "denoising_strength": 0.7
  }
}

Model Management

Switch models dynamically:

# GET /api/v1/images/models
# Returns available SD models

# Set active model
POST /api/v1/images/generations
{
  "model": "sd_xl_base_1.0.safetensors",
  "prompt": "..."
}

Image Storage

Generated and edited images are automatically stored in the file system.

Storage Flow

Generation/Edit

Image created by AI engine

Upload to Storage

Converted to standard format (PNG/JPEG)
Assigned unique file ID
Metadata stored (prompt, model, parameters)

Database Record

File entry created with:

User association
Generation parameters
Timestamp

Chat Linking

Optional: Link to chat message for context

Accessing Images

# Get image by ID
GET /api/v1/files/{file-id}/content

# Returns image file
Content-Type: image/png

Permissions

Control who can generate images:

# User permissions
{
  "USER_PERMISSIONS": {
    "features.image_generation": true  // Per-user setting
  }
}

Image generation can be resource-intensive. Consider setting usage limits or restricting to specific user groups.

Best Practices

Prompt Engineering

Write effective prompts:

Be specific about style and details
Include lighting and mood
Specify artistic techniques
Use negative prompts to avoid unwanted elements

Quality vs Speed

Balance performance:

More steps = better quality, slower
Larger sizes = more detail, higher cost
Test with low steps first
Increase for final generation

Model Selection

Choose appropriate engine:

OpenAI: Best quality, easiest setup
Gemini: Good balance, Google ecosystem
ComfyUI: Maximum control, self-hosted
AUTOMATIC1111: Free, extensive models

Cost Management

Control expenses:

Monitor API usage
Set generation limits
Use local engines for volume
Cache common generations

Troubleshooting

Connection failed to image engine

Check:

Base URL is correct and accessible
API key is valid (if required)
Firewall allows connection
Engine is running and healthy
Authentication credentials correct

Poor image quality

Solutions:

Increase step count
Use larger image size
Refine prompt with more details
Try different model
Adjust sampling parameters

Generation timeout

Fix:

Reduce image size
Lower step count
Check engine performance
Increase timeout settings
Use faster sampler

ComfyUI workflow fails

Verify:

Workflow JSON is valid
Node IDs correctly mapped
Required models installed
ComfyUI API accessible
Workflow tested in ComfyUI first

Advanced Features

Auto-Size Support

Some models support automatic sizing:

{
  "IMAGE_SIZE": "auto"  // Model chooses optimal size
}

Auto-size is only supported by models matching the pattern: gpt-image-*, dall-e-*

Batch Generation

Generate multiple variations:

{
  "prompt": "A beautiful sunset",
  "n": 4  // Generate 4 variations
}

Image Caching

Generated images are cached based on:

Prompt hash
Model configuration
Generation parameters

Repeated identical requests return cached results instantly.

API Reference

Generate Image

POST /api/v1/images/generations

Create new images from prompts

Edit Image

POST /api/v1/images/edit

Modify existing images

Get Models

GET /api/v1/images/models

List available image models

Get Config

GET /api/v1/images/config

Retrieve current settings

​Overview

​Supported Engines