You can not view this image.

This post was deleted or flagged for the following reasons:

  • [DELETION] The left image has harsh edges between contrasting colors that are very often throughout the image, accompanied by color anomalies. The middle character in the left image also has a second smaller tongue underneath the first, and the right character's irises could use a little cleaning up. In the right image, the right character's right eye could use a little cleaning up, and the sclera on the right of the right eye should be white. The bandages around her belly are broken, and you could just remove the floating nail and no one would be the wiser. The left character in the left image has what looks like lop ears and cat ears, and her left hand has 6 fingers (5 + hidden thumb) vs. 5 (4 fingers + hidden thumb) on the right. - knotanotherfurry -
Description

AI Models are capable of so many different styles, and it'd be great to see genners use a wider variety. This month I'm going to try to demonstrate various different styles.

The styles here are "oekaki" on the left and "tegaki" on the right. Like "low poly" these are styles born out of technical limitations. Oekaki comes from Japanese message boards that had simple on-site drawing software that was often quite limited, and Tegaki is from a specific blog that similarly allowed art comments with a limited toolset.

Both styles are well sampled on Danbooru so any model trained on a Danbooru dataset should be able to do these styles. Although these kind of styles don't have the full range of color and shading and such that other styles do, the limitations make them stand out in a crowd.

The model used was StableMond, which you can get at: https://huggingface.co/Xeno443 . I'd highly recommend any of the Mond models, they're all some of the best furry SDXL models available. StableMond is the most anime influenced so it's best suited for cartoony looks like this, while 3WolfMond-LastSDG is the most realistic and the others are in between.

Getting multiple characters in a gen like this is much easier with Regional Prompter: https://github.com/hako-mikan/sd-webui-regional-prompter . That lets you divide an image into regions and prompt each region separately, so you can create different characters and give them different traits. Another way to do it is to gen a background and then inpaint the characters in, that gives you more control but it's more effort.

The characters were randomized by using Dynamic Prompts: https://github.com/adieyal/sd-dynamic-prompts . That lets you randomize parts of prompts either by using { | } in the prompt or by creating wildcard files that it pulls randomly from (like in this case a file with a list of characters so it randomly selected from the list).

Hope something there is helpful to somebody!

Blacklisted

    There are no comments.