What is Mochi 1
Mochi 1 is an open-sourceAI video generationmodel released by Genmo, excelling in motion quality and adherence to user prompts. Mochi 1 is released under the Apache 2.0 license, supporting free use for personal and commercial purposes. The model currently offers a base version at 480p, with plans to release a high-definition version supporting 720p, Mochi 1 HD, before the end of the year, providing higher fidelity and smoother motion. Mochi 1's model weights and architecture can be found on the Hugging Face platform, and Genmo provides a hosted playground where users can try Mochi 1's capabilities for free.
Main features of Mochi 1
- High-fidelity motion generation: Mochi 1 has made significant progress in motion generation, capable of generating smooth videos, simulating physical principles such as fluid dynamics, fur and hair simulation, as well as coherent, fluid human motion, beginning to cross the uncanny valley.
- Strong prompt adherence: Mochi 1 can very accurately follow the prompts provided by users, generating video content that meets expectations. By jointly constructing videos from text and visual tokens, similar to Stable Diffusion 3, the streaming architecture uses a larger hidden dimension, with nearly four times the number of parameters of the text stream.
- Open-source architecture: Mochi 1's model weights and source code are released under the Apache 2.0 open-source license, and can be freely downloaded and used, supporting personal or commercial purposes.
- High efficiency: Mochi 1 uses Genmo's own Asymmetric Diffusion Transformer (AsymmDiT) architecture, which efficiently processes user prompts and compressed video tokens by simplifying text processing to focus on visual effects.
- Hosted playground: Genmo provides a new hosted playground where users can try Mochi 1's capabilities for free and experience the video generation process.
- High parameter count: Mochi 1 uses a 10-billion-parameter diffusion model, the number of variables used to train the model more accurately.
Technical principles of Mochi 1
- Asymmetric Diffusion Transformer (AsymmDiT) architecture: Mochi 1 uses Genmo's self-developed Asymmetric Diffusion Transformer architecture, which efficiently processes user prompts and compressed video tokens by simplifying text processing to focus on visual effects. AsymmDiT jointly constructs videos from text and visual tokens, similar to Stable Diffusion 3, but its streaming architecture uses a larger hidden dimension, with nearly four times the number of parameters of the text stream, and uses an asymmetric design that can reduce memory usage during deployment.
- Real-time video generation technology: Mochi 1 adopts Pyramid Attention Broadcast (PAB) technology, which achieves frame rates of up to 21.6 FPS and a 10.6x speedup by reducing redundant attention computations, without sacrificing video generation quality. This technology can provide acceleration for any future DiT-based video generation model, enabling it to have real-time generation capabilities.
Project addresses of Mochi 1
- Project official website：genmo.ai/blog
- HuggingFace model repository：https://huggingface.co/genmo/mochi-1-preview
- Online experience：https://www.genmo.ai/play
Application scenarios of Mochi 1
- Video content creation: Mochi 1 can be used to generate high-quality video content, including animations, special effects, short films, etc., providing a powerful creative tool for video producers and artists.
- Education and training: In the field of education, Mochi 1 can generate instructional videos to help explain complex concepts or simulate experimental processes, improving learning efficiency.
- Entertainment and Gaming: In the gaming and entertainment industry, Mochi 1 can be used to generate dynamic background videos or storylines within games, enhancing players' immersion.
- Advertising and Marketing: Mochi 1 can be used to generate attractive advertising videos, helping businesses create more appealing advertising content at a lower cost.
- Social Media: On social media platforms, Mochi 1 can help content creators generate unique video content, attracting more attention and interaction.
- News and Reporting: In the news industry, Mochi 1 can be used to generate background videos for news reports or simulate dynamic visual effects of events, increasing the appeal of the reports.