This is a very capable model with decent species knowledge, a wide variety of artstyles, as well as pretty good knowledge of generations of artstyles by decade going back several decades and it doesn't rely on negative tags to suppress features like human skin and furry elements if you want to make a human character or if you want to make a furry character.Some of my favorite positive tags for style has been "Anime Screenshot, Anime coloring, Digital artwork" which gets you a decent stylized anime image, if you want less 2d drop the anime coloring tag and or the anime screenshot for more of a 2d cartoon look. and if you want more 3d stuff you can tag for that too.You can use the default set of "masterpiece, best quality, absurdres" etc and "worst quality, low quality, low res" but I don't really think its necessary for this model.PLEASE NOTE! This is only tested in SDNext, aside from briefly loading it up in forge to verify something I've only ever used it in SDNext, if you have a hard time reproducing my images using your software of choice I cannot help you there. Maybe consider switching software?My example prompt:Positive: Cute female (fluffy:0.7), furry, Raccoon, detailed eyes, hair over one eye, Lace-up panties, looking away, newest, very aesthetic, best quality, absurdres, masterpiece, detailed face, A treacherous mountain peak shrouded in swirling clouds revealing glimpses of a forgotten temple carved into the rock face.Negative: blurry, poorly drawn, out of focus, (worst quality, low quality:1.2), simple background, 3d, bad hands, bad anatomy, (signature, watermark), sketch, drawing, extra digitsSteps: 16-30Sampler: DDIM, Euler, Euler A, DPM++ 2m, DPM++ 2m SDECFG: 4-10Resolution: 832-1536, This version does much better on the lower end while still maintaining strong high resolution results. I prefer either 1024x1024 or 1024x1536clip skip set to 1 (again, this is for SDNext, you probably need a different value if you're in A1111/forge or comfy)

IcedCoffeeIL

a lot to talk about here, This version is the result of around 25,000 steps of training and swapping out the text encoders, the end result is a much cleaner anime style. That being said it does have its issues, I believe backgrounds are possibly worse in this version, and it sometimes does some funky stuff with the prompt. But I would say it succeeds enough of the time for it to be worth using. The overall style is more flat anime shaded, but it still retains its ability to do detailed and soft shadows, for a more cinematic feel. It also has improved details for fur and furry characters. I hope you enjoy it! For a basic prompt setup I would recommend having your quality and style tags at the end of your prompt, and character and description tags up front. As always you can use tags or natural language, I prefer tags.

Files with SHA256: 92f05dc839b968663b63348c8a16108e25f5ef6fc92dd5e4813cade1ec0eca79