I originally ran this to teach myself a bit more about LoRA training settings. But it turned out good enough that I decided to upload it. It's trained on twelve images of a giant 50-inch tall teddy bear plush toy manufactured by a company called Sugar Tubs. But it can also be used to plushify characters by balancing the strength of their own LoRA against the weight of this one.V 1.5 info:I've found that dialing down the strength of a character LoRA, but then using ((brackets)) or (emphasis:1.5) to crank up specific keywords for character details can help with getting a likeness nailed down.It can also be used to generate non-bears, by just using plushie as a keyword, and leaving bear and teddy_bear out of your prompt.Base Model is YiffyMix_V34PONYXL info:For the PonyXL version, I simplified the tags, so that now only the unique tag T3DDYB34R is required to activate the model. If you look at the tags for the four example images appended to this version of the model, you can see how species and fur color can be modified to suit various characters and species. You can also add the tags Photograph and Photorealism to pump up the realism.Illustrious info:I ran the same set through that was used for the PONYXL version, same images, same tags, nothing changed. I only ran one pass through it, but it works surprisingly well with only that one pass. Lowering the strength drives it away from photorealism and towards (no, really?) and more illustrated look. But lower it too far and you just end up with anime girls with bear ears. I may go back and try running the same number of passes the PONY version got at some point.

Files with SHA256: aec641b38fe521d1a0c6984bef0364d08b3a69047a8264abaae1baec4a3f95dd