Prompt Trick: Extract an Image's Style as JSON to Recreate It in ChatGPT

alexcovo_eth · x · 2026-09-11

heyrobinai shares a dead-simple ChatGPT workflow: upload a reference image and prompt the model to extract its visual style as reusable JSON (colors, lighting, typography, composition, textures), keep the 3 defining traits, save it as a style profile, then generate with it and iteratively refine by comparing output vs. reference. The hardest part, per the author, is finding an image you like.

Original post →

More from Multimodal

Multimodal channel →