Bug Fixes and System Updates | Generated by AI
Fixed! Here’s what was wrong and what I fixed:
Issues found & fixed
1. _generate_scenes didn’t pass api_key — The scene generation (LLM call to create titles/subtitles/prompts) called _openrouter_chat without the API key, so it always fell back to the env var. Added api_key param through the whole chain.
2. No default LLM model for video generation — generate_video_from_content had no fallback model when model=None. Added default to openrouter/auto-beta (same as the content generation endpoint).
3. Import race condition in LocalGPUProvider — Multiple parallel threads tried to import diffusers simultaneously. The first ones hit AttributeError: module 'diffusers' has no attribute 'FluxPipeline' due to a partial import race. Fixed by pre-loading the model on a single thread before launching the thread pool.
4. Missing ffmpeg on remote — The video assembly needs ffmpeg. Installed it.
Usage
# Open the frontend
open https://rc-26cc80c1b835b7a3.radeon.firstdg.ai
# Paste your OpenRouter API key in the 🔑 field, click Check
# Select 🖥️ Local GPU (FLUX.1-schnell is pre-selected)
# Enter a topic, Generate ✨, review content, then Generate Video 🎬
Generated by AI. Curating and sharing still takes effort.
If you find it useful, feel free to
donate.
WeChat: @lzwjava
·
X: @lzwjava ·
Say hi 👋