Does the shot satisfy continuity and narrative purpose?
Review character, viewpoint, motion, composition, and relation to adjacent shots—not beauty alone.
Hobby · 2025–26
Multi-shot AIGC Creation Workflow
AIGC POV
A character, viewpoint, and constraint framework for multi-shot generative video that reduces blind retries and makes failure reasons reusable.
Why I started
Generating one attractive shot is not the hardest part. Preserving the same person, viewpoint, and world rules across shots is. Without structured records, every failure becomes another blind retry and costs keep accumulating.
The real question
How can the subjective feeling that “this shot is wrong” become specific, editable issues in character, camera, continuity, and narrative function?
Choices I made
Every action should have a reason.
Lock identity, viewpoint, and invariants before scene-specific action.
Separating global rules from local change reduces the need for the model to reinterpret the character in every shot.
Split prompts into camera, subject, environment, continuity, and negative constraints.
A modular structure maps revisions to specific variables and makes it possible to track which changes improve results.
Record failures by consistency, motion, composition, and narrative function.
A vague similarity judgment cannot guide the next round. Failure types make retries purposeful and reusable as rules for later shots.
How the system moves
How a shot moves from creative intent into controlled iteration
Start / inputProcessDecisionOutcome
- Creator
Define the shot’s narrative job
State what information, emotion, or action this shot must advance.
- Constraint layer
Lock global invariants
Fix character identity, first-person viewpoint, world rules, and continuity.
- Prompt DSL + model
Organize variables and generate
Separate camera, subject, environment, and negative constraints while recording model parameters.
Lock the shot and reference assets
Carry identity, composition, and parameters into the next shot.
Revise global constraints
Adjust character references, world rules, and invariants before retrying.
Revise local shot variables
Change camera, action, pacing, or negative constraints and regenerate.
Creative loop
Turn failure reasons into rules for the next round
Record why a candidate failed, what changed, and whether it improved; accepted shots continue to the next scene or final edit.
Business meaningThe flow replaces blind retries with explicit creative decisions: accepted shots become assets for the next scene, while failed shots identify which variable should change and why.
What I made
From judgment to something that runs.
Constrain
Define character and world rules, including what cannot change in each shot.
Generate
Create candidates from structured prompts while recording model and parameters.
Review
Select by continuity and narrative purpose, then revise constraints for the next round.
What the work left behind
- Completed a multi-scene POV film of about 1m40s
- Developed a reusable prompt DSL
- Retained a generation–review–retry record
This is a personal workflow validated on one finished piece, not yet a general tool across models and creators.
Turn review criteria into shot-level records so future work can reuse why a retry happened, not just the prompt text.