Ai Anime Arena

English | 日本語

Video.mp4

Explore the Web »

AI-Anime-Arena is an anime art benchmark platform featuring anonymous and randomized battles. All the code in this repository is hosted at https://evaluate.sh.

Important

Currently in beta; this project may be terminated without notice.

Features

• Limited to Anime-Style Images: We evaluate models used solely for anime-style artwork, not those designed to generate photorealistic or realistic images.
• Multilingual and Cross-Lingual Support: Simply copy multilingual text and paste it into the input box. No need to worry about the language.
• Comparison Including Workflows: We focus on integrated workflows that include not only pure model comparisons but also LoRA, prompt control, etc.
• Accepting Requests: You can request to add models to this repository.

Request to Add Workflows

• Please create a repository that meets the following requirements and submit a request via issues.
  • After verification, it will be automatically linked as a submodule to this repository.
• If you do not wish to make it public, you can also send a tarball or private repository to email.
  • In that case, it will not be linked as a submodule to the repository.

Requirements

• It must be an architecture that uses models focused on anime-style artwork, not designed to generate photorealistic or realistic images.
• Generation must be possible with a Python script in the form python main.py --mode 1or2 --prompt "XXXXXX" --save_path "XXXX.webp".
  • mode: Specify the image size as 1 or 2
    • 1: Aspect ratio of 3:2 (768x512, 1824x1248, etc.)
    • 2: Aspect ratio of 2:3 (512x768, 1248x1824, etc.)
  • prompt: The text used for generation
    • Strongly recommend English as it will be automatically translated within the server
  • save_path: The path to save the image
    • Recommend including the extension, preferably PNG, JPG, or WEBP
• Setup instructions must be clearly stated.
  • For how to write, please refer to this repository.
• Important: Inference speed must be within around 20seconds on a graphics card equivalent to RTX 4090.
  • Including loading time due to model size, it must be within 30 seconds.

Data Usage

• Currently under preparation

Disclaimer

• All code and datasets are released under the CC BY-SA 4.0 license. For details, please refer to LICENSE.
  • However, regarding the generated images within the dataset, please comply with each model's license.
• Please refer to the laws of your region regarding DMCA and other relevant laws.