May 19, 2025 19:00:00

China's bilibili development team announces 'AniSora,' an open source AI tool that can automatically generate animated videos

A team at

bilibili , a Chinese video sharing service, has announced AniSora , an AI tool that creates animations from a single image.

Try AniSora: The Ultimate Anime Video Generation Model
https://komiko.app/ja/video/AniSora

IndexTeam/Index-anisora · Hugging Face
https://huggingface.co/IndexTeam/Index-anisora

AniSora: Exploring the Frontiers of Animation Video Generation in the Sora Era
https://arxiv.org/html/2412.10255v2

AniSora is an open source animation video generation model developed by bilibili. The official website states, 'AniSora is bilibili's open source gift to the anime world as part of the Index-AniSora project. AniSora enables one-click movie creation for a wide range of styles, including series, Chinese original anime, manga, VTuber content, and anime PVs.'

The AniSora page looks like this. AniSora is included in Komiko , an anime and manga AI platform. Komiko also has AI tools that can be used for free by simply registering an account, but you need to sign up for a paid plan to use AniSora.

To use AniSora, simply upload the image you want to animate, enter the prompts, and select 'Anisora' from the AI models. With this, you can quickly create an animation from your image.

Komiko shows some examples of using AniSora. For example, below, from an image of an anime character running (left), a movie of a character running with his limbs waving is generated with the prompt 'A person runs forward at high speed, appearing

slightly blurred by the speed.'

In the following example,

a movie of the continuation of the dance is generated from the images of 3D model characters. The prompt specifies, 'As the camera zooms in, five girls raise their left hand above their head and lower it to knee height while dancing,' and in the movie, not only the characters but also the camera actually move slowly.

In addition, animated comics can also be generated by creating lip-syncing and character movements from manga frames as shown below.

Anisora is an open source project, more details can be found on Hugging Face and GitHub .

According to the bilibili team, although there are already several natural video generation models in the AI market, there are major challenges in video generation in the animation field, such as unique styles, exaggerated movements, and expressions that ignore the laws of physics, and the technology is still insufficient. AniSora uses over 10 million high-quality data samples for model training and introduces a controllable generative model to support multiple key animation production functions, such as image-to-video generation, frame interpolation, and local image guidance.

Although bilibili is an open source project, the samples published on GitHub and in papers use clippings and characters from Japanese anime, so care should be taken when using it.