A finetune, roughly aligning NoobAI v-pred 1.0 to Anzhc's EQ-VAE-B3 improvements? More detailed gens, less noisy generations due to cleaner latents at training time, usual v-pred shenanigans apply, but model should be more resilient to colors exploding and colors should be a bit cleaner overall compared to base v-pred, does this behave like a slight aesthetic tunning? yes, this was trained for barely 550k steps on 2x4090 I slammed my head hard against the dataset and compute wall, go donate to Anzhc for his efforts training the vae if you like the results and want him to continue improving on it, and bother mmmmmmmm in the Furry Diffusion server if you want to contribute to the Aligning efforts, we estimate this might need 2 to 4 epochs in full booru dataset, which ranges from 500 usd to 700 per epoch depending on compute provider.
If you like it, and see some future for it, donate to me at Ko-fi as this was trained at my own expense and it took several days of tries with DeepSpeed on a 2x4090 setup
Generation settings? Default v-pred, just use the bundled VAE, or B3 from Anzhc's Repo, yes he is banned, yes I'll claim the NoobAI1.1EQ he did and the VAEs here in civitai as he has given my permission to claim those.
Does this still leak random styles with some tokens due to Natural Language? yes it fucking does, fuck Natural Language with Clip L and G
I'll finetune EPS1.1 over Anzhc's previous Aligning and it will be released as well.
Some shitty comparisons
Raw gen
levels changed to expose noise in the background which remains very consistent
Same image in base v-pred
levels changed to expose noise in the background which remains very bad