animagine xl 3.1

Introduction

Animagine XL 3.1 is an advanced text-to-image generation model focusing on anime-style images. It is an evolution from Animagine XL 3.0, offering improved image quality and a diverse character range. Built on Stable Diffusion XL, it is designed for anime enthusiasts and creators.

Architecture

• Developer: Cagliostro Research Lab in collaboration with SeaArt.ai.
• Type: Diffusion-based generative model.
• Features: Enhanced understanding of hand anatomy, concept interpretation, and prompt processing.

Training

Animagine XL 3.1 was trained using 2x A100 80GB GPUs over approximately 15 days. The training involved:

• Pretraining Stage: Utilized 870k images.
• Finetuning Stages: Focused on aesthetic datasets to refine style and rendering.

Guide: Running Locally

1. Installation:

   pip install diffusers transformers accelerate safetensors --upgrade
   

2. Running the Model:

   import torch
   from diffusers import DiffusionPipeline
   
   pipe = DiffusionPipeline.from_pretrained(
       "cagliostrolab/animagine-xl-3.1", 
       torch_dtype=torch.float16, 
       use_safetensors=True, 
   )
   pipe.to('cuda')
   
   prompt = "1girl, souryuu asuka langley, neon genesis evangelion, solo, upper body, v, smile, looking at viewer, outdoors, night"
   negative_prompt = "nsfw, lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract]"
   
   image = pipe(
       prompt, 
       negative_prompt=negative_prompt,
       width=832,
       height=1216, 
       guidance_scale=7,
       num_inference_steps=28
   ).images[0]
   
   image.save("./output/asuka_test.png")
   

3. Cloud GPUs: Consider using services like Google Colab or AWS for cloud-based GPU support to optimize performance.

License

Animagine XL 3.1 is released under the Fair AI Public License 1.0-SD. Key points include:

• Modifications must be shared with the original license.
• If network-accessible, provide source code access.
• Distributions must adhere to the same or similar license terms.
• Compliance issues must be resolved within 30 days to maintain the license.