Uptade: Straight leap to 2.5 when the previous version was ready but this is version is a considerable upgrade. New architecture, should be pretty compatible with a lot of models, but if you're having troubles, consider using CFG Rescale or Dynamic Thresholding. Pretty saturated as it is of now, I'll instead keep releasing the trained artists as separated LoRAs to go along. A few style prompts have changed too, to start, the old "Blender 3D" is now the "Unreal Engine" tag, and "Blender 3D" should include a wide variety of 3D animations.

Keep in mind it thousands of images were used, not all of them were treated, so you're prone to get watermarks, for that, you can simply add "watermark, artist name" to the negative prompt.

Negative prompt should be kept simple unless you want things to change drastically with the use of embeddings, just watch out in case they override the desired style on positive prompt.

All in one checkpoint for 3D like gens. Focus on stylization and lora compatibility, good for any character traits and aesthetic. Also a personal merge I've been using for a while on some of the works you can check on my pixiv, very reliable and comfortable to work on img2img and inpainting.

Try on the most popular styles like: 3D CGI, blender, disney pixar, overwatch style, fortnite style, 3D Cartoon, mix and match to find your personal sweetspot, it's pretty responsive to style prompts. Very flexible for clothing/hairstyle/bodyweights and supports NSFW very well.

Sampling: Euler A/ Up to 30 steps.

CFG: 6~9.

Upscaling: Latent/Ultrasharp or any other GAN with with very low denoising.

VAE: DogeVAE.

The examples are not cherrypicked and all made with plain character lora, you can do a lot better and be a lot more creative.