Meet Mochi-1 – the latest free and open source AI video model

The generative AI wars are heating up as more companies release their own models. Generative videos seem to be the biggest battleground right now and Genmo is taking a different approach.

The company releases its Mochi-1 model as a “research preview,” but the new generation video model is subject to an Apache 2.0 license, making it open source and able to be disassembled and reassembled.

This also means that Mochi-1 is free to use and you can try it out for yourself on Genmo’s website. The beauty of it being open source also means that it will be available on all the usual generative AI platforms in the future and could one day run on a good gaming PC.

It’s launching into a very competitive market with various services offering a range of features, including templates from Haiper, realism from Kling or Hailuo, and fun effects from Pika Labs and Dream Machine. Genmo says its focus is on bringing the state of the art to open source.

Genmo releases free AI video model

(Image credit: Genmo Mochi-1/AI generated)

So why choose the Genmo model over all others on offer right now? It’s all about the movement. We spoke to Paras Jain, CEO of Genmo, who explained that motion is a key metric when benchmarking models.

“I think basically for a long time the only uninteresting video was one that didn’t move. And I felt like a lot of AI videos suffer from this ‘live photo effect,'” he explains. “I think our historical models had that, so the technology had to evolve. But above all, videos about movement were the most important thing we invested in.”

This first release is a surprisingly small, 10 billion parameter transformer diffusion model that uses a new asynchronous approach to pack more punch into a small package.

Jain said they trained Mochi-1 exclusively on video, rather than the more traditional mixed video, image and text approach. This gave him a better understanding of physics.

The team then worked to ensure the model could properly understand what people wanted to do with it. He told us, “We also invested very, very heavily in immediate compliance and just following what you say.”

Genmo Mochi-1

(Image credit: Genmo Mochi-1/AI generated)

Genmo hopes Mochi-1 can offer “best-in-class” open source video generation, but currently videos are limited to 480p as part of the new research preview released today.

As Jain mentions, great emphasis was also placed on quick compliance and recognition. Genmo compares this to a Vision language model as a judge following Open AI’s DALL-E 3.

Will you test Mochi-1? Let us know. It certainly faces a crowded landscape, but its open-source nature could see it expand further than some of its competitors.

It’s not even the only open source AI video model hitting the market this week. AI company Rhymes has hired Allegro as “a small and efficient open-source text-to-video model.” It’s also available with an Apache license, although at 15 frames per second and 720p, instead of Mochi-1’s 24 frames per second and 420p.

Neither model will run on your laptop yet, but as Jain told us, the beauty of open source is that one day someone will tweak it to run on less powerful hardware and we’ll be making videos offline.

Recent Posts