Mochi 1: video generator — what it does and what you need to run it

Mochi 1 is an open-source text-to-video generator created by Genmo. The tool is designed to synthesize videos from prompts and is available for local running and fine-tuning.

What it does

The authors describe the model as one of the best among open solutions for video generation. Its primary task is creating videos from text descriptions. The developers provide a tool for fine-tuning (LoRA) on your own video data, allowing the model to be adapted to specific styles or objects.

What you need to run it

The code is written in Python and distributed under the Apache-2.0 license. The model weights are also available under the Apache-2.0 license. The diffusers library is used for operation. The full set of weight files takes up 124.3 GB, with the largest file being 37.4 GB.

As described by the author, running and fine-tuning requires a GPU with 80 GB of VRAM, with H100 or A100 accelerators listed as examples. The authors do not describe running locally on less powerful hardware.

The source code is available in the repository: github.com/genmoai/mochi. The model weights are hosted at: huggingface.co/genmo/mochi-1-preview.

Who it is for

The tool is aimed at researchers and developers who have access to server-grade GPUs with large memory capacities. The fine-tuning capability makes it interesting for those working with proprietary video datasets who want to adapt the model to narrow tasks.

Mochi 1 is unlikely to suit users with consumer graphics cards: the VRAM requirements stated by the developers significantly exceed the capabilities of mainstream GPUs. The tool is also not intended for quick video generation without your own infrastructure — the authors do not mention a cloud version.

Mochi 1 offers the research community an open codebase and weights for video synthesis tasks. The high VRAM threshold limits the user base but provides room for experiments with fine-tuning on specialized hardware.