patrickz
Before you start
A spreadsheet or coding agent, three fictional shots and approved reference art.
Why this lesson exists
The original project treats AI video like software: every shot has an ID, a prompt version and a review status, and only approved references carry forward. This lab builds that register by hand before any provider integration.
The idea this guide practices: I turned a weekend into a Phantasy Star music video. Prefer a story? Read the comic Black Belt With Wi-Fi.
Do the exercise
Set your boundary
Track shot ID, purpose, prompt version, reference files, output file and review status. Start with manual generation rather than provider integrations.
Start from a production pack
Start from a production pack: a written canon and three shot intents (the guide Plan an AI video that stays consistent shows how).
Create a shot register
Create a shot register with one row per shot: ID, purpose, prompt version, reference files, output file and status.
Name files by shot and version
Name every file after its shot and version so each output traces back to its prompt.
Use four fixed statuses
Use a small fixed set of statuses: draft, in review, approved, rejected.
Carry only approved references
Carry only approved references into the next shot.
Group the register by status
Show the register grouped by status so a collaborator can find the next action.
Next sessions, not today
- Provider integrations Connect a generation provider only after the manual register works. If automation retries a paid job after a timeout, inspect its existing job ID before submitting again.
A prompt to adapt
Replace the bracketed parts with your own details.
Help me set up a file-based shot register for [my short video idea] as [a spreadsheet or a JSON file]. First version only: three fictional shots with shot ID, purpose, prompt version, reference files, output file and review status, generated by hand. No provider integrations or automatic retries. If this needs any setup, list it before creating files. Then work in six checkpoints and, after each, tell me how to check it: 1. A short written canon and three shot intents. 2. One register row per shot. 3. A file name for every output that includes its shot and version. 4. Four fixed statuses: draft, in review, approved, rejected. 5. Only approved references carried into the next shot. 6. A view grouped by status so a collaborator can find the next action.
When your first version is done, paste it and ask:
Review this shot register. List shots missing an approved reference, outputs that cannot be traced to a prompt version, and the next action for each shot.
Run this experiment
Revise shot two while preserving its earlier prompt and output. Confirm shot three points to the approved reference, not a rejected version.
Check your result
Use evidence from your output. A confident explanation from the AI is not enough.
If it isn’t working
If shot three still points at a rejected or unversioned file, stop generating. Rename every output with its shot ID and version, then point shot three at the approved reference. A good-looking output isn’t approved until you mark it approved.
Optional. Progress stays in this browser.
Where this came from
Public project repository ↗. The practice lesson is an adaptation, not a verbatim transcript. About the sources.
Prepared September 2026. Tools and interfaces change; use current official setup instructions. Session lengths are estimates.