01

Separate the stable layers

Keep brand character, style language, format behavior, palette, and user meaning in separate blocks. A new user brief should change content without silently changing the visual system.

02

Version the final instructions

Record both the template version and the selected style-recipe version. Hash the final composed prompt, not just the form fields, so a generated asset can be traced to the exact instructions it received.

03

Treat output as unverified

A probabilistic image model can still misspell text or bend a palette. Mark new work unverified until OCR, glyph comparison, palette checks, or a human review promotes it. Prompt control reduces drift; validation catches it.