Many forget that when you switch from SD 1.5 to SD XL, you also have to change the CLIP coding.

This WF was tuned to work with Magical woman - v5 DPO | Stable Diffusion Checkpoint | Civitai.

Seemingly a trifle, but it definitely improves the image quality. It avoids duplication of characters/elements in images larger than 1024px.

You will ask why as much as 4096px when the images are just over 1024.... Well, I noticed that by forcing CLIP to a higher resolution, the images generated are more consistent and more accurate.

Additionally, we still have the text_g and text_l split here. The vast majority of WFs I've come across have simply routed the same promt into both paths. Obviously, this method works :) however, if you push only the quality and style description in text_l, and the scey and quality description in text_g, the resulting images, in my opinion, are more interesting.

I use this WF to generate CHARACTERS, that's why I have a randomiser for hairstyles and face shapes plugged in, you can easily detach this and plug in a prompt directly from the ClipTextEncodeSDXL.

If you do not want to use FaceDetailer, kSampler Cycler, simply click CTRL+M on:

Have a lot of fun with this WF!

Support Paweł Tomczuk on Ko-fi! ❤️. ko-fi.com/aderek514 - Ko-fi ❤️ Where creators get support from fans through donations, memberships, shop sales and more! The original 'Buy Me a Coffee' Page.

Visit my DeviantArt page: Aderek - Hobbyist, Digital Artist | DeviantArt