beecho01/bfl-openai-image-proxy

Black Forest Labs makes a family of AI image generation models called FLUX, but their API does not speak the same language as OpenAI's image generation endpoint. Many self-hosted tools, including the popular Open WebUI chat interface, are built to talk to OpenAI-style APIs. This proxy server sits in between: you send it an OpenAI-formatted image request, and it quietly translates that into a FLUX API call, waits for the result, and hands back a response in the format the calling tool expects. The server runs as a small FastAPI application you deploy with Docker. It accepts POST requests at the standard OpenAI images endpoint, maps size values like "1024x1024" to the width and height fields FLUX expects, submits the job to Black Forest Labs, and polls the FLUX API until the image is ready. When it is, the proxy downloads the image immediately and either encodes it as base64 inline in the response or saves it locally and returns a URL pointing to its own server. The download-immediately step matters because FLUX image links expire after ten minutes, so any delay would leave the calling tool with a dead URL. The proxy lists all supported FLUX model names at a /v1/models endpoint, supports friendly model aliases, and includes a /health endpoint for monitoring. Bearer token authentication for the proxy itself is optional. By default nothing is logged, so prompts and generated images stay private. Temporary image files are cleaned up automatically after a configurable time window. The scope is deliberately narrow: text-to-image generation only. FLUX supports editing, in-painting, fine-tune management, and other operations, but this proxy does not implement them, because Open WebUI's image generation feature only needs the basic text-to-image path. Installation is either a pre-built Docker image from GitHub Container Registry or a direct Python setup. A Black Forest Labs API key is the one external dependency.