Hatsu Preview1: An Anime Model from Scratch in 1.5 GPU Days
This is a status report demonstrating the capabilities and shortcomings of Hatsu Preview1. The code and a more comprehensive report will be provided upon proper release.
Model Summary
Hatsu Preview1 is a 117M + 63M parameter tags-to-image generation model. It is created with the main objective of setting a simple baseline for anime models. The model consists of a 117M parameter image generator and a 63M parameter condition generator.
The image generator operates in a 56x80 RGB space, while the condition generator creates 1024D condition vectors given a set of tags. For ease of use, the condition generator accepts arbitrary subsets of the tags instead of requiring the user to specify every tag. As in a previous exploration, the 1024D vectors are encodings created by a pretrained tagger. The condition supports a total of ~9k rating/general/character tags, although not all of them are well-understood.
The training dataset is a 1.1M-image 1girl solo
subset of the Danbooru2023 dataset, downscaled to 56x80.
The image generator is trained for 57M images, and the condition generator is trained for 268M images. The combined training can be completed in ~1.5 days on a 4090 GPU.
Qualitative Results
Other than removing explicit images, the generated images are not cherry-picked. Each row of 4 images shares the same condition.
Basic Tags
general, dress
smile, blush, sensitive
sensitive, large breasts, cat ears, long hair, purple eyes, swimsuit
sensitive, black leotard, sideboob, blue eyes, blue hair
general, white pantyhose, black hair, medium hair, maid, detached collar
general, bench, black hair, black socks, blush, book, brown eyes, closed mouth, from above, holding, holding book, kneehighs, legs, looking at viewer, on bench, open book, pink shirt, shirt, short hair, short sleeves, sitting, skirt, smile, socks, t-shirt, thighs, white skirt
general, bench, black hair, black socks, blush, book, brown eyes, closed mouth, from above, holding, holding book, kneehighs, legs, looking at viewer, on bench, open book, pink shirt, shirt, short hair, short sleeves, sitting, skirt, smile, socks, t-shirt, thighs, white skirt
general, bare shoulders, blue background, blue eyes, blue hair, breasts, choker, cleavage, dress, facing viewer, falling petals, hair behind ear, hair between eyes, lips, looking to the side, medium hair, off-shoulder dress, off shoulder, parted bangs, parted lips, petals, ribbon, ribbon choker, simple background, skirt hold, wind, yellow choker, yellow dress, yellow ribbon
bare shoulders, between breasts, blunt bangs, breasts, brown hair, bug, butterfly, cleavage, closed mouth, dot nose, dress, field, flower, green eyes, hair flower, hair ornament, hairpin, jewelry, large breasts, long hair, looking at viewer, lying, necklace, sleeveless, sleeveless dress, sundress, thighs, very long hair, white butterfly, white dress, white flower, sensitive
`
blue dress, blush, breasts, bug, bun cover, butterfly, butterfly on hand, china dress, chinese clothes, cleavage, cleavage cutout, clothing cutout, collared dress, contrapposto, cowboy shot, double bun, dress, frilled dress, frilled sleeves, frills, gold trim, hair bun, hand up, head tilt, hip vent, lace trim, long hair, looking at viewer, medium breasts, parted lips, purple eyes, single sidelock, standing, thigh strap, thighs, twintails, very long hair, white hair, sensitive
Character Understanding
hatsune miku, general
hatsune miku, general, grey shirt, black skirt, pleated skirt
hatsune miku, sensitive, competition swimsuit
remilia scarlet, general
kafuu chino, general
kashima (kancolle), sensitive
kanzaki ranko, general
yumemi riamu, sensitive
rice shower (umamusume), general
sagisawa fumika, sensitive, cowboy shot
Image Reference
These images directly use the embeddings created by the tagger from an reference image instead of using the condition generator.
Todo
- Support non-solo images
- Scale up the model and resolution
- Add proper style and character conditioning