One prompt to a 30-second explainer video: 4 rules that make code-drawing agents reliable
_haiderfarooq · reddit · 2026-10-28
The author built a skill where a coding agent (Claude Code; also runs in Codex and Gemini) draws a short explainer reel entirely in code—paper-cutout props, a mascot, word-popped captions; one prompt in, mp4 out, no video model or API keys, 1 minute laptop render. What made it reliable: voice is the clock (local kokoro TTS + whisper.cpp word timestamps cue every scene); the agent must look at its own output (dumps 3–4 stills at prop-landing moments before export, with seeded deterministic renders so fixes stay fixed); checks it can't argue with (a script flags overlaps and mandates moving props rather than shrinking captions, which the agent otherwise does to game the check); and the skill file encodes per-scene rules, not drawing code—style fingerprints beat example code. Caveat: first render is never right, 2–3 rounds of "move that" are normal; macOS only.
More from coding & agent
- N-queens solved entirely in Python type annotations with ty, each solution a tuple — charliermarsh · 2026-10-28
- Adding Push to MCP: A New Attack Surface and a Design Sketch — braelyn_ai · 2026-10-28
- MCP gap: services can't push to agents, dev argues spec needs server-initiated events — braelyn_ai · 2026-10-28
- Swarms Cloud revamps Models page aggregating 2,000+ LLMs with pricing — KyeGomezB · 2026-10-28
- A new open source model: closed to pull requests, open to remix — kieranklaassen · 2026-10-28
- CC Switch tutorial: rotating multiple Claude official subscriptions and ban-risk notes — dotey · 2026-10-28