Only three projects under ~/claude have ever produced a video file: zerostudio (434 rendered files), content-automation (5), and an n8n hackathon landing page (2 hand-made demo reels, not AI). Ranked by reuse value for Manufactured Love.
~/claude/zerostudio, last commit 2026-02-16)Script to shotlist to voice-over to AI clips to animatic to final MP4. Next.js 16 studio UI plus .mjs CLI pipeline. Five finished episodes in output/ with final/video.mp4, animatic/animatic.mp4, per-shot visuals/sources/sNNN.mp4, 1008 rendered audio files, and manifest.json / shotlist.json receipts per run.
Providers: fal-ai/sora-2/text-to-video/pro as default video model (now deprecated on fal; FAL_VIDEO_MODEL env overrides), fal-ai/nano-banana-pro, fal-ai/gpt-image-1.5, fal-ai/fast-sdxl for stills; ElevenLabs eleven_v3 / eleven_multilingual_v2 for VO, music, SFX; Gemini for reference shotlists; Anthropic/OpenAI via AI SDK.
Stack: TypeScript, @fal-ai/client, @elevenlabs/elevenlabs-js, Inngest, ffmpeg-static, Remotion (packages/remotion-templates), Vercel Blob, Playwright.
Reusable:
scripts/fireship/generate-clips.mjs: fal queue submit/poll with retry, per-shot duration bucketing, applyNoTextGuard prompt suffix.scripts/coldfusion/finalize-video.mjs, scripts/eeaao/assemble-from-local.mjs, build-animatic.mjs: ffmpeg concat and mux.scripts/*/voice-audition.mjs: voice casting loop; maps to casting producer voices.scripts/fireship/critique-script.mjs, critique-shotlist.mjs, qa-compare.mjs: LLM QA loop on shots.scripts/coldfusion/build-timeline.mjs: VO-timing-driven cut planner.shot-renderer.tsx, kinetic-text-shot.tsx: captions and lower thirds..env.local holds fal, ElevenLabs, Gemini, OpenAI, Anthropic keys (values not read; ElevenLabs validity unverified).~/claude/image-mcp-repo/image-mcp/examples/music-videos, tools last modified 2026-01-11)Python storyboard factory: audio to Deepgram transcript plus Librosa beats to timed sections to 12 to 40 shot plan to per-shot image prompts to hero frames. Ten projects. Image and storyboard stages proven; video stage scaffolded only (projects/*/videos/trimmed/ empty).
Reusable: tools/create_storyboard_plan.py (106 KB; beat-aligned planner with per-character wardrobe and appearance locks), tools/generate_prompts.py, tools/build_timeline.py, tools/build_sections.py, tools/build_storyboard_html.py (review contact sheet), projects/la-la-land/docs/render-queue.md (shot table with per-shot model notes and continuity constraints). Port the character-lock idea into bible/characters.md and storyboard/shots.csv; do not port the Python.
~/claude/luxin-repo/luxin, last commit 2026-08-21)CLI-first hosted creative runtime across image, video, audio with credit pricing and a queue substrate. Generic creation API; no stitching, no rendered clips committed.
Reusable: docs/references/sources/*-openapi.json and *-llms.txt.md (captured request schemas for about 20 fal video endpoints, captured 2026-06-17 and 2026-08-12), docs/decisions/0026-video-modality-first-slice.md (fal video is queue-only; fal.run/<endpoint> times out), src/providers/fal.ts queue submit/poll/result substrate (8.8k lines), src/aspect-ratio-intent.ts, src/credit-pricing.ts (per-second video cost model).
~/claude/viralvlogs, 2025-08-03)Veo 3 via Vertex AI behind Inngest with a base-provider/registry/router abstraction. Unit tests and an e2e script exist; no committed renders; HEAD commit "Start to implement 'real' video generation." Superseded by Veo 3.1 on fal, which needs no GCP allowlisting. VIDEO_GENERATION_CRITIQUE.md (21 KB) is a useful self-critique of multi-provider video design.
~/claude/content-automation, 2022-08-16)oclif CLI Reddit-to-YouTube bot: crawl, script fragments, TTS, Remotion render, upload. No AI footage. Reusable as a Remotion composition pattern (src/remotion/RedditStory.tsx, Root.tsx), VO segmentation (src/processor/voice-over.ts), YouTube upload (src/commands/upload.ts).
~/claude/gemcast, 2026-01-25)src/app/(chat)/api/video/render/route.ts: SRT builder plus ffmpeg drawtext burn-in to MP4 to Vercel Blob, Node runtime, 300s max. Reference for serverless caption burn-in only.
~/claude/muskemon)scripts/gen_audio.py: ElevenLabs SFX and music from prompt dictionaries. Prompt-bank pattern for stings (confessional sting, verdict hit, fanfare).
~/claude/Google_Veo_3_Setup_Guide.md (2025-05-31) describes Veo 3 as 8s/720p allowlist-only on Vertex. Out of date.
openclaw (ffmpeg for Discord voice, not generation), image-mcp core server (image endpoints; video only as catalog entries), nobg, styl-my, taliya, vi, eros, cineguessr-repo, oscars, penguin, sundae, daily-outputs, lighthouse, legion, n8n-hackathon-2025-05-31.
Fork zerostudio's script to shotlist to VO to clips to ffmpeg spine into pipeline/. Replace its shot planner with a character-locked storyboard table. Swap Sora 2 for the Phase 5 bake-off winner. Pull request schemas from luxin's captured sources. Use the fal queue API only.