Animagine XL 4.0
New Version of Popular Anime Model for Image Generation
2 min read
Cagliostro Research Lab has released Animagine XL 4.0 (also known as Anim4gine), a major update to the popular series of models for creating anime-style images. The model was trained from scratch based on Stable Diffusion XL 1.0 (base model for text-to-image generation) on a massive dataset of 8.4 million diverse anime images. The Opt version further improves anatomy, reduces noise levels, and enhances color accuracy. The model is well-suited for anime art enthusiasts and content creators.
Key features
- Large-scale training: Used a dataset of 8.4M anime images and about 2,650 GPU hours of training.
- High detail: Support for special quality tags and scores for precise control over the result.
- Improved anatomy: The Opt version refines character proportions and reduces artifacts.
- Prompt flexibility: A special training method based on ordered tags to preserve character and style recognition.
What you can do
- Create art: Generate detailed illustrations with favorite anime characters and original heroes.
- Run in interfaces: Use the model in ComfyUI, Stable Diffusion WebUI, or via the diffusers library.
- Control proportions: Apply recommended resolutions and special quality tags for ideal results.
Example
import torch
from diffusers import StableDiffusionXLPipeline
pipe = StableDiffusionXLPipeline.from_pretrained(
"cagliostrolab/animagine-xl-4.0",
torch_dtype=torch.float16,
use_safetensors=True,
custom_pipeline="lpw_stable_diffusion_xl",
add_watermarker=False
)
pipe.to('cuda')
prompt = "1girl, arima kana, oshi no ko, masterpiece, high score, great score, absurdres"
negative_prompt = "lowres, bad anatomy, bad hands, text, error"
image = pipe(
prompt,
negative_prompt=negative_prompt,
width=832,
height=1216,
guidance_scale=5,
num_inference_steps=28
).images[0]
image.save("./arima_kana.png")Read more
DetailsBase model: Stable Diffusion XL 1.0
Parameter count: 2.57B
Training hardware: 7 x H100 80GB SXM5
License: CreativeML Open RAIL++-M
Features: requires using tags instead of natural language










Comments