Description

This PR introduces a new image provider openai_compatible that allows users to connect to any OpenAI-API-compatible image generation endpoint (e.g., LiteLLM, vLLM, Azure OpenAI, or self-hosted gateways). This feature decouples image generation from the text LLM provider, offering greater flexibility for enterprise and self-hosted deployments where image generation might be handled by a separate proxy or service.

Key Changes

• New Provider: Added openai_compatible to the ImageProvider enum.
• Configuration: Introduced new environment variables to configure the provider independently:
  • OPENAI_COMPAT_IMAGE_BASE_URL
  • OPENAI_COMPAT_IMAGE_API_KEY
  • OPENAI_COMPAT_IMAGE_MODEL (Strictly required, no default)
• Implementation: Implemented generate_image_openai_compatible method in ImageGenerationService, utilizing the standard AsyncOpenAI client with a custom base_url.
• Validation: Added strict validation to ensure all required configuration parameters (URL, Key, Model) are present before attempting generation.
• Tests: Added tests/test_image_generation_openai_compatible.py with comprehensive unit tests for provider selection, successful generation, and error handling for missing config.
• Documentation: Updated README.md with the new provider options and a specific "Using an OpenAI-Compatible Image Provider" configuration example.

How to Test

1. Run the backend tests: uv run pytest servers/fastapi/tests/test_image_generation_openai_compatible.py.
2. Start the server with the following environment variables:

   IMAGE_PROVIDER=openai_compatible
   OPENAI_COMPAT_IMAGE_BASE_URL=https://api.openai.com/v1 # or your proxy URL
   OPENAI_COMPAT_IMAGE_API_KEY=sk-proj-....
   OPENAI_COMPAT_IMAGE_MODEL=dall-e-3

3. Generate a presentation and verify that images are generated using the specified endpoint.

Checklist

• Code compiles and runs
• Tests added/updated (test_image_generation_openai_compatible.py)
• Documentation updated (README.md)
• [~] Dependencies updated (not applicable)

Closes #381