continuity-chain

name	continuity-chain
description	Generate multi-scene AI video sequences with better visual continuity by chaining each scene from the last frame of the previous clip and resuming cleanly when some scenes already exist.

Continuity Chain

Use When

• A short uses multiple generated video scenes that should feel like one evolving world.
• You are making AI-generated visual sequences rather than isolated clips.
• A prior chain run was interrupted and should resume without losing the successful scenes.

Core Rule

• If scene-to-scene continuity matters, generate scenes as a chain, not as independent prompts.

Inputs

• first seed image or first clip
• ordered scene list
• default chain prompt or per-scene prompts
• output directory for generated clips

Outputs

• ordered generated clips
• resume-safe scene inventory
• continuity notes: which scene seeded which next scene

Workflow

1. Seed scene 1 from the chosen image or initial clip.
2. For each later scene, extract the last frame of the previous clip.
3. Use that extracted frame as the input image for the next generation.
4. Skip scenes that already rendered successfully unless regeneration is required.
5. Keep the chain order explicit so a failed scene does not corrupt the rest of the run.

Good Uses

• motion-led mythology or fantasy sequences
• faceless explainers with connected generated visual worlds
• AI-image-to-video chains where abrupt style resets would look cheap

Bad Uses

• scenes that intentionally need strong visual discontinuity
• generated clips with no sequential narrative relationship
• silently continuing a chain after a failed scene without noting the break

Pair With

• Use after shot-prompt-builder when several prompts belong to one continuous generated sequence.
• Pair with partial-regeneration to resume interrupted scene chains.

Aggregated From

• digitalsamba/claude-code-video-toolkit chain_video.py
• image-to-video continuity practices from generated short pipelines

Validation Checklist

• Each scene's seed source is recorded.
• Existing successful scenes are skipped safely on resume.
• Failed scenes do not get silently treated as valid chain inputs.
• Visual continuity is materially better than independent scene generation.