oh-ashen-one

12 Hour YouTube Videos — Data skill for Claude Code

Data community

Ambient YouTube video pipeline: AI images → FFmpeg → 12h sleep/lofi videos.

How to install 12 Hour YouTube Videos

This entry records only its repository, not the path inside it, so there is no exact command to give. Open oh-ashen-one/12-Hour-Youtube-Videos and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What 12 Hour YouTube Videos does

Ambient YouTube video pipeline: AI images → FFmpeg → 12h sleep/lofi videos.

Alternatives in Data

README

12-Hour YouTube Videos 🏮

**Feed this repo to your AI agent (Claude Code / Codex / Kimi / OpenClaw / etc.) — e.g. copy it into your agent's skills directory or paste the SKILL.md — and it can run an automated ambient YouTube channel: generate one AI clip + one AI track, loop them into an 11h55m video with FFmpeg, and upload it to YouTube Studio for ~$0.00–0.10 per video.**

Everything in here is distilled from running this pipeline in production — including every way it broke.

The Key Insight

You only need **ONE 8-second AI video clip** (Veo) and **ONE ~3-minute AI audio track** (Suno). FFmpeg loops both to fill up to 11h55m. The render is free CPU time. A month of daily content costs under $2.

1 Veo clip (8s) + 1 Suno track (~3min)
            ↓
   FFmpeg loops both → 11h55m video
            ↓
   ffprobe verify → re-encode to 1.5–2.0GB
            ↓
   YouTube Studio upload (browser injection)

What's Inside

  • SKILL.md — the canonical skill: both pipelines (Veo-clip generation + loop render, and production render + automated YouTube Studio upload), setup gates, cost guardrails, content strategy, lessons learned, failure catalog, best practices
  • scripts/ — render scripts and thumbnail generator (see table below)
  • references/ — deep-dives the SKILL.md routes to:
    • scenes.md — copy-paste Veo prompt catalog by theme
    • suno-prompts.md — proven Suno prompt formulas
    • ffmpeg.md — all render / re-encode / verify commands
    • youtube-upload.md — the browser-injection upload method, step by step

Hard-Won Rules (the short version)

  • Render at -t 42900 (11h55m) — YouTube silently rejects ≥12:00:01
  • ffprobe-verify duration on the exact file before upload — no exceptions
  • Raw renders are 5–19GB → re-encode -crf 35 -maxrate 300k to 1.5–2.0GB before browser upload (CRF 28/32 still fail)
  • Upload via local CORS file server + DataTransfer injection; the selector is **input[name=Filedata], NOT `input[t