Skip to content
alexjercanPublic

About

Experiment to see how well Claude can do YouTube slop

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Agent YouTube

Create your own niche and try to beat my stats!

A YouTube Shorts studio run end to end by an AI agent. The agent picks the subjects, writes the simulations, renders the footage, speaks the narration, composes the music, draws the art, and prepares each upload. Every narrated claim is measured by the same simulation that rendered the footage.

Niche

Simulation-driven science shorts. Each short runs one real, deterministic simulation and makes one surprising claim about it: probability paradoxes, chaos, emergence, and algorithms in motion. The claim is never written from memory. The simulation measures it, the task records the number, and the narration quotes the measurement. The seed appears on screen, so every video is reproducible by anyone.

Why this niche fits this agent:

  • The footage is code. The agent writes correct, deterministic render code.
  • The claims are computable. Accuracy is enforced by assertion, not by vibes.
  • The visuals are native to the format: motion, loops, and a payoff frame.
  • No camera, no gameplay license, no stock footage, no copyright exposure.

See docs/niche.md for pillars, audience notes, and the idea backlog.

Channel identity

  • Name: Seed Zero
  • Handle: @SeedZeroLab (@SeedZero was taken at creation time).
  • Tagline: "Real simulations. Measured claims. Seed on screen."
  • About text: "Every video on this channel is a real, deterministic simulation. Every claim you hear was measured from that exact run. The seed is on screen, so you can reproduce it. Written, rendered, voiced, and scored by an AI agent."
  • Avatar and banner: generated in-repo (tracked as a branding task).

Initial channel settings

Setting Value
Channel type Brand Account under the owner's Google account
Category Education
Default language English
Audience Not made for kids (channel-level)
Altered content disclosure Yes; the channel is openly AI-made
Default visibility Private on upload; the agent publishes after its own QA pass (owner authorized 2026-08-30)
Comments On, hold potentially inappropriate for review
License Standard YouTube license
Country Owner's choice

Format and cadence

  • Vertical 1080x1920, 60 fps, VP9 or H.264 source, 30 to 45 seconds.
  • Voice: standalone Piper API, piper-1, en_US-lessac-medium, with direct whisper.cpp verification.
  • Music and sound: procedural, generated in-repo, no licensed audio.
  • Cadence: target 3 shorts per quota day, released after the 10:00 local quota reset. Quality gates may reduce the day's count.
  • Shorts only. Other platforms (TikTok, IG Reels) reuse the same vertical file if the owner grants access later. Long form is out of scope.

Layout

  • docs/ — vision, niche, and channel setup guides.
  • sims/ — deterministic simulation and render code, one folder per short.
  • projects/ — per-video production data: script, manifest, metadata.
  • media/ — rendered output, ignored by git.
  • scripts/ — production helpers (voice-over, music, composition).
  • systemd/ - user service and timer for the daily production run.
  • tasks/ — Tatr-tracked work.
  • web/ - static status page: stats, the video and idea slate, and a work log. Data lives in JSON and JSONL under web/data/. Run python -m http.server -d web to preview it locally. Pushes to master deploy it to GitHub Pages through .github/workflows/pages.yml.
  • secrets/ — OAuth credentials for uploads, ignored by git.

Quickstart

scripts/voiceover.sh projects/<name>/narration.txt media/<name>/voice.wav

The script uses Piper at http://localhost:10303/v1/audio/speech and whisper.cpp at http://localhost:10301/inference. Set TTS_API or STT_API to override either full endpoint.

Daily production run

scripts/seedzero-produce runs Claude in this repository with Fable at xhigh effort and invokes /daily-production. It streams progress, requires the final PRODUCTION_STATUS: COMPLETE line, and writes each report under ${XDG_STATE_HOME:-~/.local/state}/seedzero/.

scripts/seedzero-produce
scripts/seedzero-produce --open
scripts/seedzero-produce --detach
scripts/seedzero-produce --open-last
scripts/seedzero-produce --logs

--open runs production, then opens that run's report. --detach starts the run in a transient user service. --open-last opens the newest saved result.html without starting a run. --logs follows the scheduled seedzero-produce.service with journalctl --user -fu. Use only one mode at a time.

Install the command and enable the 10:00 local systemd timer:

./install.sh

Run the launcher tests with:

python3 -m unittest scripts/test_seedzero_produce.py

Channel access

The agent needs upload and analytics access to work autonomously. The owner setup steps are in docs/channel-setup.md.

About

Experiment to see how well Claude can do YouTube slop

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages