Mochi 1is an open source AI video generation model from Genmo that converts text prompts into high-quality video. It is released under the Apache 2.0 license and represents an important milestone in the democratization of AI video technology, supporting free use for personal and commercial purposes. The model is currently available in a base version at 480p, with plans to release a high-definition version, Mochi 1 HD, with 720p support by the end of the year, offering higher fidelity and smoother motion.The model weights and architecture for Mochi 1 are found on the Hugging Face platform, and Genmo provides a hosted playground where users can try out Mochi 1's features for free. What makes Mochi 1 special is its 10 billion (10B) parameter architecture, making it the largest open source video generation model ever released. The model excels in two key areas: cue compliance and motion quality, setting a new benchmark for what is possible in open source AI video generation. As a free and easy-to-use tool, Mochi 1 breaks down the barriers that have traditionally kept advanced AI video generation capabilities out of reach for many creators and developers.
Mochi 1 Features
- High Fidelity Motion Generation: Mochi 1 makes significant strides in motion generation, generating smooth videos that simulate physical principles such as fluid dynamics, hair and hair simulation, as well as coherent, fluid human movements that begin to cross the Valley of Horror.
- Powerful cue following: Mochi 1 follows user-supplied cues very accurately, generating video content that meets expectations. Video is built through the joint construction of text and visual markup, similar to Stable Diffusion 3, with a streaming architecture that has almost four times as many parameters as a text stream through a larger hidden dimension.
- Open Source Architecture: The model weights and source code of Mochi 1 are released under the Apache 2.0 open source license and can be freely downloaded and used to support personal or commercial use.
- Efficient: Mochi 1 uses Genmo's own Asymmetric Diffusion Transformer (AsymmDiT) architecture, which efficiently handles user prompts and compressed video tokens by simplifying text processing to focus on visuals.
- Hosted Playground: Genmo offers a new hosted playground where users can try out the features of Mochi 1 for free and experience the video generation process.
- High number of parameters: Mochi 1 uses a diffusion model with 10 billion parameters for the number of variables used to train the model more accurately.
Official website link:https://mochi1.me
