Heritage Structure — Clay Doc Shorts Pipeline
19 scripts and a set of gates for clay stop-motion documentary shorts. Rules enforced in code, not written in a guide.
A working pipeline for clay stop-motion documentary shorts about buildings and civil engineering. Nineteen scripts and a set of gates — not a PDF telling you what to do.
Why gates instead of a guide
A guide is read once and forgotten by episode four. These rules are enforced in code, so the pipeline refuses to continue when you break them.
The hook gate rejects encyclopedia openings. "Hwaseong Fortress was built by King Jeongjo" gets swiped away. "Here is a fortress built from brick instead of stone" makes someone ask why. The gate can tell the difference and it will stop you.
The research gate catches figures and dates with no source behind them.
The narration gate is the one that surprises people. Text-to-speech mispronounces technical terms and Chinese characters silently — the waveform looks perfect and every structural check passes. So the pipeline synthesises the narration, listens back to it, and compares against the script. On one episode it caught four mispronunciations out of six terms. There is no way to find those by looking.
Cut timing is fitted to measured audio length, not to an estimate from character count. Estimating is how a five-minute plan becomes four minutes twenty.
The upload step blocks publishing when the playlist, the synthetic-media disclosure, tag length, hashtag count, category or thumbnail is missing. Each of those was forgotten at least once before it became a gate.
What is inside
Nineteen engine scripts: clay-look prompt generation, still and clip ordering, baked-text removal, pin-and-label annotation overlay, narration synthesis, listen-back verification, selective re-synthesis of only the wrong lines, audio-fitted retiming, mixing with ducking, sound effects, final assembly, contact-sheet review, and a YouTube uploader that refuses incomplete metadata.
Five documents: the pipeline doctrine and its narrative skeleton, a setup guide with a verification command at every step, a full walkthrough of one episode start to finish, a failure log of what credits and hours were spent learning, and a topic list with the criteria for choosing.
Being straight with you
The bundled channel configuration is Korean — Typecast Korean voice, Malgun Gothic subtitles, a Korean prose gate. The look, the gates, the timing logic and the upload discipline have nothing to do with language, but running this in another language means changing the voice and font in the channel config yourself.
The setup guide is written for Windows, using PowerShell and winget, and the default subtitle font path is a Windows path. It runs elsewhere; you will edit the font path.
You need paid accounts. Image and video generation runs on Higgsfield at roughly 320 credits per episode. Korean narration runs on Typecast, which is a subscription. Sound effects, narration verification and YouTube OAuth work on free tiers.
The sample episodes shown are from the Japanese-language sister channel. Same engine, same gates, different channel configuration.
This is not a button. Topic choice and script are yours. What the pipeline guarantees is that the look, the structure, the timing and the checks come out the same every time.
The visuals are AI-generated. Platform disclosure requirements — and, if you sell or advertise with them in Korea, AI labelling law — are the buyer's responsibility.
No promises about views or revenue.
Requirements
Python 3.11 (3.12 and above break some packages) · ffmpeg · a Higgsfield account with credits · a Typecast key for Korean narration · free-tier keys for ElevenLabs, Gemini and Google Cloud.
Licence
Use it on your own channels with no episode limit. Redistributing or reselling the code and documents is prohibited. Delivering to third-party clients across multiple channels requires a separate studio licence. Copyright in the videos you make is yours; the generation services' own terms still apply to you.
Lifetime updates.
—
Where this sits
- The Prompt Atlas — $49 — the corpus the pack's grammar was derived from — 1,297 shipped prompts
- Six Months of Spend — $29 — what running this standard actually cost: 2,464 clips, itemised
- The whole catalogue — teardowns, references, skill packs and free calculators, with what each one is for