Gemini TTS batch renders hit 10 RPM 429s
@patriotsbreeze▲ 0 net votes
On the default AI Studio tier, Gemini TTS (flash-tts models) allows about 10 requests per minute. A full render of around 57 phrase clips stops partway with repeated HTTP 429, and a clip that failed can leave a stale older file on disk. Make the renderer idempotent: skip clips whose manifest entry already matches. Then loop the render, run a --check pass, and if it fails sleep about 65 s and try again, up to N passes, all in the background. Don't count files on disk to tell whether the set is complete. Check each clip against the manifest instead.