GitPedia
rushindrasinha

rushindrasinha/youtube-shorts-pipeline

Automated YouTube Shorts pipeline: news → script → AI visuals → voiceover → captions → upload

1 Releases
Latest: 1w ago
v3.1.0 — community providers, CJK captions, reliability fixesv3.1.0Latest
rushindrasinharushindrasinha·1w ago·June 9, 2026
GitHub

📦 Merged community contributions

  • CJK caption rendering (#13, @chimaek): niche profile `captions.font_family` and `captions.font_size` are now wired into the generated ASS style, so Korean/Japanese/Chinese and other non-Latin scripts render with the right font instead of boxes.
  • MiniMax provider (#12, @octo-patch): optional LLM (chat completions) and TTS (streaming) behind `MINIMAX_API_KEY`. No new dependencies; ships with mocked tests.
  • 60db TTS provider (#20, @Dev-develope): optional TTS behind `SIXTYDB_API_KEY` with native Indic-language voices, per-niche voice settings, and a `python -m verticals voices --provider 60db` listing command.

🐛 Fixes

  • `edge-tts` pin bumped to `>=7.0,<8.0`. The 6.x client is rejected by Microsoft's endpoint with HTTP 403, which broke the default free TTS path. Verified against the live endpoint.
  • Missing `GEMINI_API_KEY` no longer surfaces as a cryptic `403: Method doesn't allow unregistered callers` (#11): b-roll falls back with a clear message, thumbnails fail fast, and 403 errors hint at AI Studio vs Vertex credential type.
  • ffmpeg builds without libass no longer crash assembly; caption burn-in is skipped with a warning and the SRT still uploads.
  • `SKILL.md` and `references/setup.md` now use `--topic` consistently (#14) and the `python3 python -m verticals` typo is gone.
  • TTS fallback chain is now edge → minimax → elevenlabs → 60db → say.

📝 Docs

  • New `examples/` directory with two verified end-to-end walkthroughs: a full tech-news Short and a zero-cost Ollama draft with topic discovery (#8).
  • README now documents all three Claude auth paths — API key, `claude login`, and `CLAUDE_CODE_OAUTH_TOKEN` from `claude setup-token` — and the latter is now accepted as a CLI credential source for headless use (#5).
  • Full changelog: https://github.com/rushindrasinha/youtube-shorts-pipeline/blob/main/CHANGELOG.md