Keep a character or subject consistent across multiple image generations
Each generation is independent by default. Consistency has to be engineered in.
- Use it for
- Anyone generating a series of images meant to feature the same character or subject.
Image generation treats each prompt independently by default, so asking for "the same character" across multiple images without engineering consistency in produces a family resemblance at best — similar but visibly different people. Real consistency requires locking down and reusing a specific, detailed description of the subject's actual features, not a vague label like "a young woman with brown hair" that leaves too much for the model to reinvent each time.
This prompt builds that detailed, reusable description first, before generating anything.
When not to use this
If visual variety between images is fine or even desirable, this level of lock-down isn't necessary — let each generation be independent. This is specifically for when the same recognizable subject needs to appear across a series.
Did this work?
Across several generated images, the same specific character features (not just "similar vibe") recur — recognizable as the same subject, not a family resemblance.
Tested on claude-opus-5. Evidence status is draft; it moves to battle-tested only on recorded runs, never by hand.