multiclip-continuity
Skill0 starsChanged 2 months agoMITView on GitHub
Use before delivering ANY multi-clip AI video script, and after every round of prompt revisions - audits consecutive clip pairs for spatial, prop, wardrobe, cast, knowledge, time and payoff breaks, then produces a regeneration ledger saying exactly which clips to re-generate and which to keep. Model-agnostic; applies to Veo, Omni Flash, Seedance, Wan, Kling and anything else that renders one clip at a time.
In this repository9 other files it gives its agents.
- cinematographySkill
- dialogue-and-voiceSkill
- identity-and-likenessSkill
- prompt-expanderSkill
- seedance-promptingSkill
- veo-flow-promptingSkill
- veo-prompt-lintSkill
- viral-short-scriptSkill
- wan-3-promptingSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Every clip-at-a-time generator has the same blind spot: it never sees your other clips.
- Distinctive: line 40
Every character referenced in ACTION or DIALOGUE must exist in that clip's LOOK line.
- Distinctive: line 63
After every round of prompt changes, state exactly what to regenerate — in chat
- Distinctive: line 79
Every CHECK and REGENERATE verdict carries a one-line reason plus a "how to verify" anchor — a phrase the correct clip should visibly …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Multi-clip continuity and the regeneration ledger
Every clip-at-a-time generator has the same blind spot: it never sees your other clips. Continuity is not something the model does badly — it is something the model cannot do at all. It is your job, and it is mechanical.
This skill is model-agnostic. Use it with
veo-flow-prompting,
wan-3-prompting or
seedance-prompting.
1. The continuity protocol (mandatory before delivery)
Audit every consecutive clip pair and the whole story. For scripts beyond ~8 clips, run the audit as a separate pass (or a separate agent) rather than inline — it is where tired eyes fail.
1. SPATIAL Locations connect. Door and prop sizes stay consistent (an object hiding a door must itself be door-sized). Clip N's camera entry matches clip N−1's exit. When characters enter a room, put the camera inside, facing the doorway.
2. PROPS Track possession per clip: who is holding what, right now. Stow items before two-handed actions (pocket the loot before climbing). Object scale can never change — a fist-sized stone cannot "become" a necklace; make it a pendant from the first reveal.
3. WARDROBE Restate outfits in every clip or the generator invents new ones. Dishevelled states must progress logically — nothing un-tears, nothing un-soaks.
4. CAST Every character referenced in ACTION or DIALOGUE must exist in that clip's LOOK line. An eyeline to "him" without him in the LOOK generates a random stranger.
5. KNOWLEDGE Characters cannot act on plot facts they do not know. The deceived partner cannot perform insider spycraft two clips before the reveal.
6. TIME AND AUDIO BRIDGES Time of day stays consistent. Sound bridges adjacent clips (an orchestra faint through walls; rotors approaching before the helicopter appears). Vehicles must arrive and depart on screen — no teleports between clips.
7. SETUP AND PAYOFF Planted lines and motifs must pay off on screen. If a clip says "midnight", something happens at midnight.
2. The regeneration ledger (mandatory after every revision round)
The operator generates clips while the script is still changing. Without a ledger, every prompt edit reads as "regenerate everything", which is false and expensive.
After every round of prompt changes, state exactly what to regenerate — in chat and as a dated status card at the top of the script document. Update the card each round; do not stack old ones.
| Verdict | Meaning |
| KEEP | Prompt unchanged this round (or only labels/notes changed). |
| CHECK | Only reinforcement lines changed — wardrobe restated, fidelity line, audio flavour. Keep the existing clip if it already shows those details correctly; regenerate only if it visibly diverges. |
| REGENERATE | Material change: action, dialogue, camera geometry, prop or possession, who is in frame, or a continuity fix a neighbour depends on. |
| GENERATE | New, or not yet generated. |
Rules:
- A change in clip N does not force regenerating N+1…end. Only clips whose own prompt changed, or whose existing footage now visibly breaks against the new neighbour.
- Every CHECK and REGENERATE verdict carries a one-line reason plus a "how to verify" anchor — a phrase the correct clip should visibly contain — so the operator can tell which prompt version their footage came from.
- The status card mirrors the chat table 1:1: same verdicts, same reasons.
Example row:
Clip 06 — REGENERATE. The pendant now exists from clip 03, so the reveal gesture changed. Verify: the correct clip shows the pendant already at her throat before she touches it.
3. Naming and ordering
Name every generated clip Clip NN - <5-6 word title> (e.g.
Clip 06 - The reveal at the doorway). Generate in story order. Keep the canonical
titles in one place — the script document — and use the same titles in the ledger, the
filenames and the edit. Version drift between "the clip I generated" and "the prompt I
think it came from" is the most common way a day of generation gets thrown away.
4. Where the copy lives
The script is the source of truth and it lives in the repository, not in a scratch directory and not only in a chat window. Anything served for review is a throwaway mirror of the repo file. Scratch directories get reaped between sessions; the work disappears with them.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

