Named after the pioneering chronophotographer Étienne-Jules Marey, the Marey model will provide filmmakers with an ethical alternative to existing generative AI models, many of which have been criticized for scraping publicly available content without permission. “We used all our resources, and a vast network of creators, to make sure we have a licensed, curated generative AI model,” said Asteria CEO Bryn Mooser. This effort was supported by XTR, the documentary studio and streaming platform Mooser also founded, which played a key role in securing clean training data for the model.
Unlike consumer-grade generative AI video models that focus on casual content creation, Marey is built for high-end production environments. “Our thesis is that to make production-grade generative video viable, the key missing piece is control,” explained Moonvalley CEO Naeem Talukdar in a recent appearance on the AI/XR Podcast. “Today’s video models are mostly text-to-video—type in a prompt, and you get a clip. That’s nowhere near what’s needed for professional filmmaking.”
Moonvalley CEO Naeem Talukdar
Marey provides granular control over camera movement, object motion, and environmental consistency, which are critical for filmmakers looking to integrate AI into professional workflows. It also incorporates motion capture and video-to-video transformation tools, allowing creators to direct AI-generated content with the precision expected in traditional production settings. Asteria has been working with a range of directors and artists, including Paul Trillo, to refine these tools in real-world creative scenarios.
Moonvalley, which recently secured $70 million in funding, has assembled a team of top AI researchers from DeepMind, Google, and Meta to develop Marey. Talukdar emphasized that the team is committed to overcoming the challenges of training high-quality video models on limited, legally sourced data. “Most AI models are trained on an ocean of unlicensed content, which gives them an edge in variety but creates legal uncertainty. We’re doing it differently, ensuring every pixel Marey has seen is accounted for.”
BEVERLY HILLS, CALIFORNIA – JANUARY 07: (L-R) Natasha Lyonne and Bryn Mooser attend Netflix’s 2024 … [+]
Despite these advances, Marey remains in early testing. A waitlist of creators is eager to test its capabilities, with Asteria and Moonvalley selectively granting access as they refine the tool’s outputs. The goal is to ensure that Marey delivers high-resolution, cinema-quality video without the technical pitfalls seen in other generative models, such as loss of coherence between frames or upscaling artifacts.
Marey enters a competitive landscape, where established players like OpenAI’s Sora and Google DeepMind’s Veo are pushing the boundaries of AI-generated video. However, the distinction of a fully licensed training dataset sets Marey apart. “This is about building an AI company that Hollywood can trust,” said Talukdar.
If you can’t tell the difference, there is no difference.
Adobe, another major player in AI video tools, has also taken steps to ensure ethical AI implementation with its Firefly Video Model, which integrates directly into Premiere Pro. Firefly allows users to generate and enhance video clips with commercially safe AI. However, it has received mixed reviews, with some users disappointed by its limitations and pricing model. While Firefly’s focus is on refining existing footage, Marey’s ambition is to generate entirely new cinematic content from scratch.
As AI video technology continues to evolve, Asteria and Moonvalley’s approach reflects a growing demand for transparency and accountability in generative media. Whether Marey will become a staple in Hollywood productions remains to be seen, but its development signals a shift toward more ethical AI adoption in the entertainment industry.