<p>Trained on Skidoo's art style. Because the style changed over time I tried to train the older and newer style. The older sketchier style is invoked with oldski, the newer cleaner style with newski.  The Training Data is not images, but prompts for the images.  The images can be somewhat recreated with the prompts.</p>

Skidoo

<p>Skidoo art style, trained on V2 image set.  Using the tPonynai3 v6.5 model. I used DPM++ SDE and a strength of 0.7-0.8. Use keyword oldski for the older sketchier style, and newski for the newer cleaner style. Trained this with a slightly larger artist image set and altered captions, so captions from the V1 lora might not work. Prompts will also have to be somewhat modified if used directly from the artists e621 tags. The zip included in the coconut24 V2 lora has captions it was trained on, for examples of keywords and so people can see it's capabilities.</p>

Files with SHA256: f11831d56da89d0f1394fbd8a9c92ce5e5ebf5b053af5577d2a50e8aa8dd0fc7