Trained on 184 images of Coyote (no repeats) from Kemono Friends on Anything v3.0 Pruned FP32 (however it works on every model) This time I added the tag "coyo" to the training so that it associates "coyo" with many features of her, and fixes the association with "coyote" and "kemono (in parenthesis!)" messing up the generation Of course, coyote (kemono friends) still works Both vtuber and regular design are in the training data, but mostly vtuber. Doesn't like generating her legwear correctly, insists that both are yellow. Stats: 184 512x images (no repeats) for 19 epochs Network dimension: 32 Base model: Anything-V3.0-pruned-fp32.ckpt Batch size of 3 on RTX 3060 Generation tags: coyo, 1girl, ((solo)), animal ears, shirt, short hair, beige hair, virtual youtuber, yellow gloves, yellow eyes, coyote ears, tail, beige pleated skirt, coyote girl, coyote tail, white sleeveless crop top, jacket, white necktie, green boots, white belt, white choker, spaghetti strap, bare shoulders, long sleeves, navel, midriff, hair between eyes, collarbone, open jacket, asymmetrical legwear, american flag, dark skin, bra strap, paw print soles, shirt writing, dark brown jacket Sample images are 512x512 latent upscaled to 1024x1024 and generated on Anything v3.0 Pruned FP32 |