How to do img2img

[arrowxpr]

I just started using SwarmUI and editing an existing image has been kind of confusing. I tried adding images to init and try different prompts, inpainting, etc and the results still aren't as expected.

How can we achieve better results for when we want to pick a subject from an existing image and produce new images containing subject in different shapes/styles. I hope my question is clear, thanks!

[mcmonkey4eva]

Your description isn't actually "img2img" (image initialization, often used for masked subregion editing or quality refinement), you want image prompting, see docs here https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Features/ImagePrompting.md

The most popular option for that right now is to use Qwen Image Edit, with Flux2 behind (better quality, much more performance-intensive to run). If you're very resource limited, SDXL finetunes with IP-Adapter might suffice but won't nearly match the quality of modern techniques.

[arrowxpr]

Oh, in the description I meant to frame it as generating a new image based on a previous image. For example changing hair color, or adding accessories to a subject etc. Will try those models you suggested. Thank you!

[mcmonkey4eva]

Yeah that'd be great case for edit models. Flux2 Klein just released and is the new best fast-to-run edit model, so that's the one to go for

[diantekyrie]

Yeah that'd be great case for edit models. Flux2 Klein just released and is the new best fast-to-run edit model, so that's the one to go for

Will this be in a later version or is there a plugin/extension for us to use?

[mcmonkey4eva]

it just came out a couple days ago #1 (comment)