VVV READ THIS, EVERYTHING VVV

Trained on 106 artists and pictures from r34 (33k images total). Model can do hyper now. Cant do immobile or tubefeed stuff at all. Will be fixed on a next version coming up really soon.

Artists included is in the training data file that you can download from civitai using the dropdown menu in download.

Know issues: some output can occasionally be low res, some artists work better than others. Characters may end up holding wands if you dont prompt for arm movement such as hands on hips arms over head and such.

Recommended settings and prompt for generation:

((best quality)), (((masterpiece))), looking at viewer, etc.
Negative prompt: Photoshop, video game, ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, mutation, mutated, extra limbs, extra legs, extra arms, disfigured, deformed, cross-eye, body out of frame, blurry, bad art, bad ???????, 3d render, comic, text, speech bubble, holding, holding food, food
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 3165871059, Size: 512x640, Model hash: 5147a9f178, Denoising strength: 0.65, Clip skip: 3, Hires upscale: 1.65, Hires steps: 14, Hires upscaler: Latent (antialiased)

You should use nai vae

hires fix is a good idea to use, it works really well at these settings.

I've had help from people in my discord to put together the dataset and I'm very thankful for that, and we are going to continue working on it to make it better. Thanks.

If you got some questions ask me in my discord.

A v3.5 is coming up very soon to fix some problems.