ugc-video-adlisted

Produce or iterate a 9:16 UGC-style video ad (TikTok, Reels, Shorts) with AI personas and no video SaaS - persona, still photos, image-to-video clips through an API (for example Kling via the OpenRouter video endpoint), a deterministic HTML timeline rendered frame by frame with Playwright, then H.264 with music and synthesized sound design through ffmpeg. Includes the creative rules learned from real campaigns (nobody talks, no breathing or sighing, plain faces, a new setting and topic per video, product only at the end, three hooks per persona, caption safe zones, AI-content disclosure) and the ffmpeg contact-sheet commands to check frames before rendering. Use it whenever the user talks about a video ad, a TikTok or Reels creative, UGC video, a new variant or hook, a new persona for an ad ("a student", "a mom"), a low 6-second view rate, or says a video "looks AI", even if they don't name the pipeline.
josephbedminster/indie-marketing-skills · ★ 1 · AI & Automation · score 74

Install: claude install-skill josephbedminster/indie-marketing-skills

# UGC video ad with AI personas A 15-16 second vertical ad, produced locally for a few dollars: a persona struggles with a real problem, the product demo solves it, the persona reacts, an end card names the product. Everything runs from one variants file, so a new hook costs zero generation and 90 seconds of rendering. ## Before you start Read `.claude/ugc-config.md`: pipeline location, commands, variants file, models, cost, brand invariants, lessons already learned, results so far. Missing? Create it with the questions and template of the `ai-persona` skill (`../ai-persona/references/config-template.md`). Read **Business profile** and **Project conventions** of `.claude/posthog-funnel-map.md` if present. No pipeline yet? `references/pipeline.md` describes the architecture to build. ## Creative rules (learned the expensive way, apply by default) Each clip costs about half a dollar and two minutes, so these save real round-trips. Details and the failures behind them: `references/lessons.md`. - **Nobody talks.** No AI voice-over (it sounds cheap), and a mouth moving with no sound is uncanny. Every clip prompt says the lips stay closed. Hooks are situations (stuck on homework at 11 pm), not someone addressing the camera. - **No breathing, sighing, blowing, puffed cheeks.** Video models draw visible steam or mist: the most obvious AI tell. Forbid it in a shared clip suffix and never ask for it in a clip prompt. - **Plain faces, not models** (see `ai-persona`). - *