framepipe-dev/media-inference-worker ? reverse-engineered prompt

Build me a small Python tool that can send a text prompt to a media model inference service and print the result.

I want to be able to run something like python generate.py qwen-image-3 "Editorial portrait, hard flash, 35mm grain" and have it make the request, wait for the response, and show me the output clearly in the terminal. Please keep it simple and practical, since this is just a lightweight client for testing model generation.

Use the existing README and RUNBOOK as the source of truth for how the request should work, and if you need anything current from the provider docs, look it up online. It should handle basic errors well, like missing settings, bad responses, or request failures, and make it easy to swap in other endpoints or request formats later if needed.