← sources

What are you using for image to prompt generation ?

source: reddit (reddit:StableDiffusion)published 2026-10-28 08:38ingested 2026-10-29 02:40

retained text

The idea is to submit any type of image to a local "analyzer" to generate an *effective*\* T2I prompt.

I have a very basic ComfyUI workflow based on Qwen 3 VQA, but it isn't very apt at creating a good prompt.

By "effective" I mean: description of the scene, subjects - with pose and some facial clues, relative spatial relationships, setting. The minimum requirements to be able to re-create the source image, and then tweak it at will.

I've also tried some I2I workflows with little success; and I generally find that it doesn't offer the same level of flexibility.

document fetched 2026-10-29 00:33 · canonical https://reddit.com/r/StableDiffusion/comments/1xhilm1/what_are_you_using_for_image_to_prompt_generation