Innovate Futures @ Benji
patreon
[In Progress] Kolor IPA Virtual Try-On With Flux Img2img
🕑 Added 2024-10-16 21:44:50 +0000 UTCKolors and it's IPAdapter do really strong style transfer.
I just remember I forgot to continue with this model.
Updated the currently version, and here are some testing result.
2nd pass , Flux result
other t-shirt
Kolors 1st pass
another example, same pose
Mask
Kolors 1st pass
Flux 2nd pass
Will need to some more change. Final workflow post later , free for all.
Cheers !
Comments
Benjamin Law
Btw, there are new diffusion model from Nvidia themselves. Low parameters but processing speed faster than Flux and SD3.5 multiple times. They are not use T5 text encoder, they used LLM for the trxt encoding.
Aarika
Sure Thanks Ben
Benjamin Law
Yes, i will add segmentation in. It have SEG anything 2 and Florence 2 in there actually, I just like to use mask editor.๐
Aarika
Hi Ben, Instead of manually selecting masks it would be better to use some segmentation to detect the mask and pass it to the next node. This will create a free flow of workflow without any manual intervention.
