adrlisted

# ADR V8.1 — Automated Documentary Rendering Input a topic, output a cinematic short video with voiceover, subtitles, BGM, and social media copy. ## Execution ```bash cd /Users/wekoidubai/ADR && nohup /opt/homebrew/bin/python3 run_adr_v8.py "主题" > /tmp/adr_v8.log 2>&1 & ``` Required env: WERYAI_API_KEY, TG_BOT_TOKEN, TG_CHAT_ID. Self-reports to Telegram. Do NOT monitor or tail logs. ## Architecture ``` Topic -> Spielberg (Gemini 3.1 Flash Lite, 3x retry for 9 lines) -> Historian (Gemini 3.1 Flash, visual research) -> Jiang Wen (Gemini 3.1 Flash, image prompts) -> Voice Director (Gemini 3.1 Flash, speaker selection) -> WeryAI Podcast (single master voice track) -> Whisper ASR (base model, auto-download) -> Velocity-curve interpolation (char_time_map per sentence) -> 20-thread pool (9 AI images + 1 BGM) -> ffmpeg (concat demuxer + ASS subtitles + J-Cut + amix) -> Telegram (video + social copy + one-click copy) ``` Cinematic timing: image -> +0.2s subtitle -> +0.5s audio (J-Cut via -itsoffset) ## V8.1 Key Changes - All LLM calls unified to Gemini family (no Claude dependency) - Whisper alignment: velocity-curve interpolation replaces greedy-merge - Script generation: 3x auto-retry if < 9 lines - Env vars: graceful error + debug dump - Whisper model: auto-download