Skip Navigation

Search

SDXL images coming out weird

Has anyone seen this? Been trying to get XL to work, got the safetensors added, git pull'ed, redid the pip install command, restarted, and still get this.

Any ideas?

Thanks to @[email protected] , I had a VAE turned on, switched that to None and it works beautifully now

7

Help working with an extremely narrowly trained model

Hey folks, first time posting here, I hope this is the right place. I'm having a frustrating issue, I downloaded a Lycoris and added the extension to SD, and it works great, but was clearly only trained on close up portraits.

As soon as my prompt includes anything outside of the shoulders it switches to someone else. I tried training my own Lora with more poses to be more varied, but my face is distored now.

Can I combine these, either in prompts, or can I combine the models somehow? Could I use img2img on only the portrait area? (I tried but I don't know how to focus on just the head area, like "Apply this lycoris to the (masked) head"

Thanks!

[EDIT]: I figured it out, I was using whole picture and didn't understand how the "Only masked area" worked. I played with it a lot and figured out I could create a base image and have a wonky face, then in img2img mask the face and then bring in the lycoris, use only masked, and it'd fill it perfectly

1