Moonvalley’s Marey is a video of artificial intelligence

Photo of author

By [email protected]


Join daily and weekly newsletters to obtain the latest updates and exclusive content to cover the leading artificial intelligence in the industry. Learn more


A few years ago, there was no such thing as “AI Tawnidi Video Model”.

Today, there are dozens, including many who are able to submit a very high high video clip, or a video of real Hollywood standards in seconds of the text or images that are uploaded to the user and existing videos. If you have read Venturebeat in the past few months, there is no doubt that you coincide with articles on these models and companies behind them, from Runway’s Gen-3 to Google’s Veo 2 To Openai for a long time, but finally available Sura to Luma Aiand BeccaAnd Chinese Chinese Cling and Hailuo. until Ali Baba It called for startup Ginmo Open source video models were shown.

Indeed, these models were used to make parts of the main films, from Everything, everywhere once to The real investigator of HBO: Night Country To music videos and television ads from Our games and coca cola. But although Hollywood and relatively fast film makers of artificial intelligence are still, there is still one big problem on the horizon: copyright fears.

The best we can say, given that most of the AI ​​-startup companies do not publicly share accurate details of their training data, most of which are trained on vast areas of videos that have been uploaded to the web or collected from other archival sources, including those that have copyright that their owners may give or may not have given explicit permission to AI video companies. In reality, The runway is among the companies facing a collective lawsuit (She is still working on her way across the courts) on this particular case, and It is said Also for this purpose. The conflict continues on whether the data is bulldozing, including videos, is fairly used and converted.

But there is now a new alternative to those concerned with copyright and unwillingness to use models as there is a question mark. An emerging Moonvalley Founded by Google DeepMinders and former researchers from Meta, Microsoft and Tiktok, among other things – Marey, which is MARY, presented a combination AI video model designed for Hollywood studios, film makers and institutions brands. Marey was placed in its position as a “clean” model to establish AI, and Marey is trained exclusively on the owned and licensed data, providing an ethical alternative to artificial intelligence models that were developed using scraping content.

“People have said it was not technically possible to build a video of advanced artificial intelligence without using diaphragm data,” said Moonvalley CEO and founder of Naeem Talukdar in a recent video call interview with Venturebeat. “We have proven otherwise.”

Mary, is now available on the basis of the waiting list only, joins Firefly Adobe Video FormAnd who says the long software seller that this is in terms of institutions-after he was only trained in licensed data and Adobe stock data (to the panic of some shareholders) – Provides compensation for institutions for use. Moonvalley also provides compensation for Paragraph 7 of this documentSaying that she will defend her customers at her own expense.

Moonvalley hopes to make these features attractive for large studios – even like others like others The runway meets with them -Film manufacturers, among an endless and constantly growing collection of new video creation options from artificial intelligence.

More “ethical” video intelligence video?

Mary is the result of the cooperation between Moonvalley and Asteria, which is the AI’s film studio led by artists. The model is designed to help creative professionals instead of replacing it, and providing film makers with new tools to produce video artificial intelligence while maintaining traditional industry standards.

“Our conviction that you will not get a prevailing adoption in this industry unless you do this with this industry,” said Talukdar. “The industry was high and clear that in order to actually use these models, we need to know how to build a clean model. And even today, the upper path could not do that.”

Instead of stripping the Internet, Moonvalley built direct relationships with creators to license their shots. The company took several months to create these partnerships, and to ensure that all the data used for training and licensing them completely obtained.

Moonvalley license strategy is also designed to support content creators by compensating them for their contributions.

“Most of our relationships really come in reality after people began to hear what we are doing,” Talukdar said. “For young creators, many of their shots are only sitting. We want to help them invest in it, and we want to focus models focusing on artists. It ends with a very good relationship.”

Talukdar Venturebeat told that although the company is still evaluating and reviewing its compensation models, it generally compensates the creators based on the duration of their shots, and they pay them every hour or concern under fixed licensing agreements (for example, 12 or four months). This allows possible repeated payments if the content continues to be used.

The goal of the company is to make extremist video production easier and costly, allowing filmmakers, studios and advertisers to explore the story of the stories created from artificial intelligence without legal or moral concerns.

More cinematic control – beyond text claims, photos and camera trends

Talukdar explained that Moonvalley pursued a different approach with the Marey AI video model of the current artificial intelligence video by focusing on professional production instead of consumer applications.

“Most of the obstetric video companies today focus on the consumer,” he said. “They build simple models as they demand a chat, create some clips and add great effects. Our focus is different: What is the technology needed for Hollywood studios? What do the main brands do to make Super Bowl ads?”

Mary offers many developments in the video created from artificial intelligence, including:

  • The original HD generation A high -resolution video is born without relying on height, and reduces visual artifacts
  • The length of the extended video Unlike most AI’s video models, which only generate a few seconds of shots, Mary can create a 30 -second sequence in one pass.
  • Type -based editorial Unlike other gym video models, Marey allows users to edit the introduction, medium and background separately, providing more accurate control of video composition.
  • Inputs the story and drawing -Instead of relying only on text claims (as many artificial intelligence models do), Marey film makers can create stories, graphics and even live references, making them easier for professionals.
  • More in response to air conditioning inputs The model is designed to better explain external inputs such as graphics and movement references, which makes the video created by artificial intelligence more controlled.
  • The “original” video editor Moonvalley develops a program accompanying Mary, which works as the original video editing tool that helps users manage projects and timelines more effectively.

“The model itself is very built on the possibility of control,” Talukdar explained. “You must have more controls around the output – the ability to change letters. It is the first model that allows you to make editing based on the layer, so you can edit the front, middle and background separately. It is also the first model that has been built for Hollywood, specially designed for production.”

In addition, he told Venturebeat that Mary depends on a mixed -transfer model that combines spread and transformer -based structure.

Talukdar said: “The models are models of models scattered, so they are the structure of transformers, then you have a spread as part of the layers.” “When you offer control, usually do this through those layers you make.”

Funded by VCS with large names, but not as much as other AI video companies (so far)

Moonvalley also announced this week a seed tour of $ 70 million led by Bessamer Venture Partners, Khosla Ventures, and General Catalyst. Hemant Taneja, Samir Kaul and Byron Deeter joined the company’s board of directors.

Talukdar pointed out that the financing of Moonvalley is much lower than some of its competitors, so far – the existence of a runway has been reported Rapid 270 million dollars in a total of several rounds – But the company has improved its resources by collecting the elite team of researchers and engineers from artificial intelligence.

He said: “We raised about 70 million dollars, that is, a little less than our competitors, for sure.” “But this really summarizes the team – the presence of a team that can build this architecture more efficiently, calculating, and all of these different things.”

Marey is currently at a limited access stage, with selected studios and movie factories testing the form. Moonvalley plans to gradually expand access in the coming weeks.

“At the present time, there are a number of studios that have been able to reach, and we have the alpha group with a few dozen directors who use it,” asserted Talukdar. “Hope is to be fully available in two weeks, the worst condition in two months.”

With the launch of Marey, MONVALLY and Asteria, they aim to put themselves at the forefront of artificial intelligence filmmaking, as studios and brands offer a solution that integrates artificial intelligence without prejudice to creative safety. But with my competitors starting the self -codification like Runway and Pika and Hedra Continue adding new features such as the voice of personality and movements, the field has become more competitive.



https://venturebeat.com/wp-content/uploads/2025/03/moonvalley-marey.png?w=1024?w=1200&strip=all
Source link

Leave a Comment