Image Generation

Generate images using DALL-E, Flux, Stable Diffusion, and Gemini with AI agents

Overview

LibreChat supports multiple image generation engines that can be used by AI agents to create images based on text descriptions. Agents can generate images as part of conversations or workflows.

Supported Engines

DALL-E

OpenAI’s DALL-E 2 and DALL-E 3 models.

# .env
DALLE_API_KEY=your-openai-key
DALLE3_API_KEY=your-openai-key  # Can be separate

# librechat.yaml (optional customization)
# DALLE3_SYSTEM_PROMPT="Custom system prompt"
# DALLE3_BASEURL="https://api.openai.com/v1"

OpenAI Image Tools

Customizable OpenAI image generation for agents:

# .env
IMAGE_GEN_OAI_API_KEY=your-key
IMAGE_GEN_OAI_MODEL=gpt-image-1
# IMAGE_GEN_OAI_BASEURL=custom-url

Custom descriptions:

IMAGE_GEN_OAI_DESCRIPTION="Generate images with custom AI"
IMAGE_GEN_OAI_PROMPT_DESCRIPTION="Custom prompt description"
IMAGE_EDIT_OAI_DESCRIPTION="Edit images with AI"

Flux

Black Forest Labs Flux models for high-quality generation.

# .env
FLUX_API_KEY=your-flux-key
FLUX_API_BASE_URL=https://api.us1.bfl.ai  # or https://api.bfl.ml

Get your API key at api.us1.bfl.ai

Stable Diffusion

Self-hosted or cloud-based Stable Diffusion WebUI.

# .env
SD_WEBUI_URL=http://host.docker.internal:7860

For Docker setups, use host.docker.internal to access the host machine.

Gemini

Google’s Gemini models with image generation capability.

# .env
GEMINI_API_KEY=your-gemini-key  # Or use GOOGLE_KEY
GEMINI_IMAGE_MODEL=gemini-2.5-flash-image

# librechat.yaml (for Vertex AI)
# GOOGLE_SERVICE_KEY_FILE=/path/to/service-account.json
# GOOGLE_LOC=us-central1

Configuration

1. Choose Image Provider

   Select one or more image generation engines based on your needs:

   • DALL-E: Best integration, commercial use allowed
   • Flux: High quality, flexible licensing
   • Stable Diffusion: Self-hosted, fully customizable
   • Gemini: Multimodal integration with Google models
2. Configure API Keys

   Add the necessary API keys to your .env file:

   # .env - Choose the engines you want
   DALLE3_API_KEY=your-openai-key
   FLUX_API_KEY=your-flux-key
   SD_WEBUI_URL=http://localhost:7860
   GEMINI_API_KEY=your-gemini-key
   
3. Enable for Agents

   Image generation is automatically available to agents when configured. No additional settings needed in librechat.yaml.

Using Image Generation

In Conversations

Direct Request

Generate an image of a serene mountain landscape at sunset
with a lake in the foreground.

As Part of Workflow

Create a marketing campaign for our new product:
1. Write a tagline
2. Generate a hero image showing the product in use
3. Draft social media copy

With Specific Style

Generate an image in the style of impressionist painting:
A Parisian café scene with people enjoying coffee.

Image Generation Tool

When an agent has image generation enabled, it uses the tool automatically:

// Agent automatically calls image generation tool
{
  tool: "generate_image",
  parameters: {
    prompt: "A serene mountain landscape at sunset",
    size: "1024x1024",
    quality: "hd"
  }
}

Generation Progress

During image generation, users see a progress indicator:

// Progress tracking component
<ImageGen
  initialProgress={0.1}
  args="Creating Image..."
/>

The progress animation shows:

• Visual spinner
• Status text (“Creating Image”, “Finished”)
• Progress percentage

File Configuration

Control image generation settings:

# librechat.yaml
fileConfig:
  imageGeneration:
    percentage: 100  # Scale to percentage of original
    # OR
    px: 1024  # Fixed pixel size

Advanced Options

DALL-E Azure Integration

Use DALL-E through Azure OpenAI:

# .env
DALLE3_AZURE_API_VERSION=2024-02-15-preview
DALLE2_AZURE_API_VERSION=2024-02-15-preview

Custom DALL-E Endpoints

# .env
DALLE_REVERSE_PROXY=https://your-proxy.com
DALLE3_BASEURL=https://custom-dalle.com
DALLE2_BASEURL=https://custom-dalle.com

Stable Diffusion Configuration

For self-hosted Stable Diffusion WebUI:

1. Launch Automatic1111 WebUI with API enabled:

   ./webui.sh --api --listen
   

2. Configure the URL:

   SD_WEBUI_URL=http://localhost:7860
   

Gemini Authentication

Multiple authentication methods:

API Key (AI Studio)

GEMINI_API_KEY=your-key

Default, easiest for development.

Shared Google Key

GOOGLE_KEY=your-key

Shared with Google chat endpoint.

Vertex AI Service Account

GOOGLE_SERVICE_KEY_FILE=/path/to/service-account.json
GOOGLE_LOC=us-central1

For production deployments.

Image Specifications

DALL-E

• DALL-E 2: 256x256, 512x512, 1024x1024
• DALL-E 3: 1024x1024, 1792x1024, 1024x1792
• Formats: PNG
• Quality: Standard or HD (DALL-E 3)

Flux

• Resolutions: Up to 2048x2048
• Formats: PNG, JPEG
• Models: Flux Pro, Flux Dev, Flux Schnell

Stable Diffusion

• Fully customizable based on your model
• Common: 512x512, 768x768, 1024x1024
• Supports various samplers and schedulers

Gemini

• Model: gemini-2.5-flash-image
• Integrated with Gemini multimodal capabilities
• Supports various aspect ratios

Troubleshooting

Image generation fails

• Verify API keys are correct and active
• Check API quota/billing status
• Ensure network connectivity to image service
• Review error messages in logs

Stable Diffusion not connecting

• Verify WebUI is running with --api flag
• Check firewall settings
• Use host.docker.internal for Docker setups
• Confirm URL includes port number

DALL-E 3 prompts rejected

• DALL-E 3 has content policy restrictions
• Avoid restricted content (violence, adult, etc.)
• Rephrase prompts if rejected
• Check OpenAI status page for outages

Images not displaying

• Check file storage configuration
• Verify browser can access image URLs
• Check file size limits in fileConfig
• Clear browser cache

Best Practices

• Detailed prompts: More specific descriptions produce better results
• Style keywords: Include art style, lighting, perspective
• Iterate: Refine prompts based on results
• Multiple engines: Different engines excel at different styles
• Cost awareness: Monitor API usage, especially for DALL-E 3

Example Prompts

Good: "A photorealistic portrait of a golden retriever puppy
playing in autumn leaves, soft natural lighting, shallow depth
of field, professional photography"

Poor: "dog picture"

Related Features

• Agents
• Multimodal Chat
• Artifacts