Marie of Moonvalley’s Marey is the latest AI video model trained in fully licensed data
Join our daily and weekly newsletters for the latest updates and exclusive content of a leading AI coverage industry. Learn more
A few years ago, there was no such thing as a “generative video model of AI.”
Today, there are dozens, including many, capable of making ultra -high definition, Hollywood ultra -rowalistic video caliber in seconds of text hints or images, which is stacked with users and existing videos. If you have read Venturebeat in the last few months, undoubtedly come across articles about these models and companies behind them, from Gen-3 on the track yes Google’s I see 2 to the long-term but finally accessible to Openai Sora yes Forward., Peakand Chinese up Cling and HailS Even Alibaba and a startup, called Genmo have offered open source video models.
These models have already been used to make parts of the main blockbusters, from Everything, everywhere, everything at once yes HBO’s real detective: night side to music videos and TV ads from Toys r ‘us and Coca ColaS But despite AI’s relatively quick hug of Hollywood and movies, there is still a big potential problem: Copyright concern.
As much as we can say, considering that most start -ups for video model AI do not publicly share accurate details about their training data, most are trained in huge layers of videos uploaded to the network or collected by other archival sources, including those with the owners of which may or may not have actually explicit video companies for video companies. Actually, The track is among the companies facing a trial (still working through the courts) on this and It is reported, As for this purpose. The dispute continues whether to scrape data, including videos, represents honest and transformational use.
But now there is a new alternative for those who deal with copyright and do not want to use models where there is a question mark. Starting called Moonvalley – Founded by former Google Deepminders and researchers from Meta, Microsoft and Tiktok, among others – presents Marey, generative video model AI designed for Hollywood studios, movies and corporate brands. Positioned as a “clean” state-of-the-art fundamental AI video model, Marey is exclusively trained in possessed and licensed data, offering an ethical alternative to AI models developed with the help of cut content.
“People said it was not technically possible to build an avant -garde AI video model without using carved data,” said Moonvalley CEO and co -founder Naeem Talukdar in a recent video conversation interview with Venturebeat. “We have proven otherwise.”
Marey, now available on a waiting list, joins Adobe’s Firefly Video Modelfor which this long -established software provider says, Adobe stock data (to the horror of some associates) – and provides compensation to businesses for use. Moonvalley also provides compensation of Clause 7 of this documentSaying that he will defend his customers on his own account.
Moonvalley hopes these features will make marey attractive for Big Studios – even as others like The track concludes deals with them -And movie creators, among the countless and ever -growing array of new AI videos.
More AI’s Ethical Video?
Marey is the result of collaboration between Moonvalley and Asteria, a studio for films and animated and animated studio, led by an artist. The model is built to help, not to replace creative professionals, providing new video producers with AI -managed video production, while maintaining traditional industrial standards.
“It was our conviction that you would not get basic acceptance in this industry unless you do so with the industry,” Talukar said. “The industry was strong and clear that in order to actually use these models, we need to figure out how to build a clean model. To this day, the above song was that you couldn’t do it. “
Instead of scraping the Internet for content, Moonvalley has built a direct relationship with the creators to license their staff. The company took several months to establish these partnerships, ensuring that all data used for training are legally acquired and fully licensed.
Moonvalley’s licensing strategy is also designed to support content creators, offsetting them for their contribution.
“Most of our relationships are actually entering the incoming ones now when people have started to hear about what we are doing,” Talukar said. “For the creators of small cities, many of their footage just sit around. We want to help them provide it and want to make models focused on the artist. After all, it’s a very good relationship. “
Talukdar told Venturebeat that while the company still evaluates and reviews its compensation models, it usually compensates the creators based on the duration of their staff, paying them hourly or briefly a percentage of fixed licensing agreements (eg 12 or four months). This allows potential repetitive payments if the content continues to be used.
The goal of the company is to make high-end video more accessible and cost-effective, allowing the makers of movies, studios and advertisers to explore AI stories without legal or ethical problems.
More cinematographic control – beyond text prompts, images and camera instructions
Talukdar explained that Moonvalley uses a different approach with its video model Marey AI from existing AI video models, focusing on a professional production and not consumer applications.
“Most generative video companies are more focused on users today,” he said. “They build simple models where you drive a chatbot, generate some videos and add cool effects. Our focus is different: what is the technology needed for Hollywood studios? What do the main brands need to make ads on Super Bowl? “
Marey introduces several achievements in a video generated by AI, including:
- Native HD generation -Gires high resolution video without relying on increase, reducing visual artifacts
- Extended Video Length -In unlike most AI video models that generate only a few seconds of frames, Marey can create 30-second sequences in one pass.
- Layer -based editing – Unlike other generative video models, Marey allows users to separately edit the foreground, middle part and background, providing more precise control over the video composition.
- Stories and sketches based inputs -Instead of relying only on text prompts (as many AI models do), Marey enables the makers of films to create the use of boards, sketches and even live references, which makes it more intuitive to professionals.
- More responsive to conditioning entrances -The model is designed to better interpret external inputs such as drawings and movement references, which makes AI the generated video more controlled.
- “Generative-native” video editor -Moonvalley develops accompanying software for Marey, which functions as a tool for editing a video for generative video that helps users manage projects and terms more efficiently.
“The model itself is simply built very strongly around the controllability,” Talukar explained. “You need to have much more controls around the output – you can change the characters. This is the first model that allows you to edit based on a layer, so you can edit the foreground, the middle of the ground and the background separately. This is also the first model designed for Hollywood designed for production. “
In addition, he told Venturebeat that Marey relies on a hybrid model of a transformer diffusion that combines diffusion and transformer-based architectures.
“The models are diffusion transformative models, so this is the transformer architecture and then you have diffusion as part of the layers,” Talukar said. “When you enter controllability, you usually do it through these layers.”
Funded by VCS with big names but not as much as other AI startup videos (yet)
Moonvalley also this week announces a circle of seeds of $ 70 million, led by Bessemer Venture Partners, Khosla Ventures and General Catalyst. Investors Hemant Taneja, Samir Kaul and Byron Deeter have also joined the Board of Directors of the company.
Talukdar noted that Moonvalley’s funding is significantly less than some of its competitors so far – it has been reported that the track has Raised a total of $ 270 million for several rounds – But that the company optimizes its resources by assembling an elite team of researchers and AI engineers.
“We raised about $ 70 million, much less than our competitors, for sure,” he said. “But this really comes down to the team – the presence of a team that can build this architecture much more efficiently, to calculate all these different things.”
Marey is currently in a limited access phase, with the selected studios and filmmakers testing the model. Moonvalley plans to gradually expand access in the coming weeks.
“There are currently a number of studios that get access to it, and we have an alpha group with several dozen filmmakers using it,” Talukar confirmed. “The hope is that it will be fully available within a few weeks, the smallest case within a few months.”
With the launch of Marey, Moonvalley and Asteria, they aim to position themselves in the AI-assisted film creation by offering studios and brands a solution that integrates AI without compromising creative integrity. But with AI video startup rivals such as a track, Pika and Honoring Continuing to add new features as a voice and movement character, the field is becoming more and more competitive.