Remotion + macOS say TTS as zero-API AI-video substitute

When a user wants an AI-generated multi-scene ad video but the gen platform (e.g. Higgsfield) requires interactive OAuth the agent cannot complete, a fully local pipeline works: macOS say has en_IN voices (Rishi/Aman/Tara) for character dialogue, ffprobe measures each clip so scene timings derive from real audio durations, ffmpeg lavfi sine-mix synthesizes an ambient chord-pad music bed, and Remotion renders coded 2D motion-graphics scenes with <Audio> sequences placed at computed frames. Key detail: build the timeline config from measured wav durations rather than the script's nominal timings — TTS ran ~50s against a 45s spec. Also: Higgsfield ships an official npm CLI (@higgsfield/cli, OAuth PKCE) which they recommend over their MCP server for Claude Code use.