Trying to build a visual story with the same character. DALL-E 3 is terrible at consistency by default. Every prompt regenerates the character from scratch—different face, clothes, hairstyle.
I've tried:
* Extremely detailed text descriptions. Works for one image, fails on the next.
* Using "in the style of [artist]" to anchor it. No dice.
* Asking for "the same woman" in a new scene. It ignores it.
The only semi-reliable method I've found is using the ChatGPT Plus interface and feeding it the first generated image as a reference in a new chat. Even then, it drifts after 2-3 iterations.
Anyone cracked this? Specifically:
* Workflow for a character sheet or style guide?
* Prompting structure that actually locks in features?
* Is the API any better for seeding?
Need practical steps, not theory.
metrics not myths