FaceBooru-DiT-256x256px

t2i latent diffusion model trained on 100k images of anime faces dataset for ~80 epochs. 45hrs RTX A6000. Ended on train loss of ~0.12.

• 170M Trainable Parameters
• Text encoder is google-t5/t5-small
• VAE is REPA-E/e2e-sdvae-hf

Capabilities:

• Can generate 256 by 256px okay images
• Image features can be altered by changing tags

Limitations:

• Generated images have faded colouring (from how the vae interacted with one of the datsets)
• Difficulty generating hands, neck, collarbone correctly

Colab link at: https://colab.research.google.com/drive/1xqqUi4Re2-wGXldKfXNH64OsSDY1ZY1w?usp=sharing

Better-looking example images:

general, 1girl, solo, long hair, brown eyes, simple background, white background, open mouth, jewelry, heart, earrings, frills, choker, black hair, looking to the side, maid, maid headdress, animal ears, portrait, heart earrings, hair between eyes

general, 1girl, solo, short hair, looking at viewer, blush, frown, blue eyes, ribbon, hair between eyes, close mouth, school uniform, blonde hair, sky, day, cloud, blue sky, red ribbon, eyelashes, window, portrait, close-up

1girl,long hair,blonde hair,purple eyes,smile,rabbit hair ornament,pink ribbon,sidelocks,frills,sleeveless,heart brooch

v6 8500k steps This model has been pushed to the Hub using the PytorchModelHubMixin integration:

• Code: [More Information Needed]
• Paper: [More Information Needed]
• Docs: [More Information Needed]