Genmo, an AI startup, has announced the launch of the Mochi 1 AI video generator, showcasing the capabilities of their 480p base model. Mochi 1 is released freely under an open-source license, helping drive the AI video generation forward. Its availability will empower diverse applications. It demonstrates high-fidelity motion, strong prompt adherence and realistic video outputs. Let’s get into the details!
Table of Contents
Example Video Generated by Genmo Mochi 1
Key Features of Mochi 1
Some key highlights of this video generator include:
1. Motion Quality
The model is able to generate smooth videos at 30 frames per second for up to 5 seconds in length. The motion appears natural and physically plausible with fluid dynamics. Actions like the movements of human characters look increasingly realistic.
2. Prompt Adherence
Users have a high level of control over the generated videos thanks to Mochi 1’s strong alignment with textual prompts. Videos accurately reflect the descriptive instructions and capture requested details around characters, environments and desired actions.
3. Model Architecture
Powering these capabilities is Genmo’s novel Asymmetric Diffusion Transformer (AsymmDiT) architecture, representing over 10 billion parameters. This efficient design jointly reasons over video tokens and user prompts to generate high-quality outputs.
Experience Mochi 1 on the Genmo Playground
To experience this model directly, Genmo has launched an online Playground at https://www.genmo.ai/play. Here, users can freely generate and view videos using their own prompts without any limitations. The weights and architecture for Mochi 1 are also open and available on HuggingFace. You can also access the source code at github.com/genmo/models.
Performance Evaluation of Genmo Mochi 1
Previous open video models have fallen short on motion quality and accuracy. Mochi 1 definitively raises the bar, setting a new best-in-class standard that even outperforms some closed systems. Its strengths across key metrics establish it as the leading open-source option. While limitations remain as an early research model, Mochi 1’s abilities have been rigorously evaluated:
1. Prompt Adherence
An automated test judges how carefully videos align with given instructions, scored by a vision-language model. Mochi 1 480p preview has strong prompt adherence. It outperforms top AI video generators like OpenAI Sora and RunwayMl Gen-3, Kling, etc.
2. Motion Quality
The following ratings focus on the interest and physical realism of motions like fluid movements, assessed by human judges. Again, Mochi 1 outperforms OpenAI Sora and RunwayMl Gen-3.
Funding for the Future
Genmo has raised $28.4 million in Series A funding led by NEA to continue progressing open video AI for the benefit of all. This investment will support expanding Mochi 1’s abilities while building future visionary research. With Mochi 1, Genmo cements its commitment to enabling open collaboration in video AI.
What’s Next?
Mochi 1 HD is coming later this year, which will support 720p video generation to address edge cases such as warping in complex scenarios using enhanced fidelity and even smoother motion. Future work will focus on expanding the model’s control and resolution capabilities.
Join Genmo
As leaders in open video generation, Genmo offers exciting career opportunities for talented researchers and engineers. Candidates are encouraged to apply to help unlock new frontiers of general intelligence for the benefit of all. By sharing their research for free while continuing ambitious work, Genmo drives the entire field forward into an open future.
| Latest From Us
- Forget Towers: Verizon and AST SpaceMobile Are Launching Cellular Service From Space

- This $1,600 Graphics Card Can Now Run $30,000 AI Models, Thanks to Huawei

- The Global AI Safety Train Leaves the Station: Is the U.S. Already Too Late?

- The AI Breakthrough That Solves Sparse Data: Meet the Interpolating Neural Network

- The AI Advantage: Why Defenders Must Adopt Claude to Secure Digital Infrastructure







