Same guy working on HLL inpainting (it's uploading to HF now). On a whim, I decided to try something different: merging the instruct-pix2pix model with the chuuba model to see how it does. It does okay. You can use this to edit images in certain ways without having to inpaint (such as turning day in an image to night, or changing colors of clothing).
I've made the Pikamee in
>>44120212 sad. Note that when you're using this model, both positive and negative prompts must be instructional rather than just tags. I've provided a catbox as an example on how to do that. As you can see, I've also tried it with the original instructpix2pix model as well, and it does worse at this particular task, meaning the merging with the chuuba model does do something.
Please note that it won't easily replace one chuuba in an image with another (At least, not without multiple iterations and changing the background). Not that advanced. I'll upload the HLL3-ip2p image on HF next if people are interested.
Original:
https://files.catbox.moe/vgmmhl.pngHLL:
https://files.catbox.moe/k0npt5.pngInstruct pix2pix info:
https://huggingface.co/timbrooks/instruct-pix2pix 