Reduce the Prompt After the Visual Decisions Are Fixed
Reduce the prompt after stable visual decisions have been moved into other artifacts.
Capture
Prompts continued to grow even after identity, environment, composition, objectives, and geometry had been established.
Additional sentences attempted to protect every successful element. Some generations became less stable because the central action was buried among repeated constraints.
Why
More information felt safer.
Every sentence appeared capable of preventing another misunderstanding, and a comprehensive prompt looked more controlled than a short one.
Why-Not
The earlier decisions had already removed many uncertainties.
Repeating them did not increase control. It increased prompt surface and made the unresolved instruction harder to distinguish from background detail.
The scar was learning that a mature workflow should make prompts shorter, not longer.
Commit
Decision: After visual decisions are fixed, remove every prompt instruction that merely restates them.
Retain only the information required for the new motion, performance, camera behavior, or change.
A prompt should direct unresolved decisions rather than memorialize resolved ones.
This produces a more controlled generation, but it does not yet produce knowledge. When a clip fails, the workflow must learn which remaining decision caused it.
Timestamp
2026-07-24