TLDR: Los Angeles-based AI startup Moonvalley has launched Marey, a ‘3D-aware’ video synthesis model uniquely trained 100 percent on public domain films. This initiative directly addresses widespread ethical and legal concerns surrounding AI models trained on proprietary data, offering a ‘clean model’ approach to generative AI.
In a significant development for the artificial intelligence landscape, Moonvalley, a Los Angeles-based AI startup, has introduced its groundbreaking video synthesis model, Marey. What sets Marey apart in the often-contentious field of video-generating AI is its foundational training: the model is reportedly 100 percent trained exclusively on public domain films. This approach directly challenges the prevailing industry norm where complex algorithms are frequently fed proprietary material, leading to extensive ethical and legal debates over data sourcing.
Video-generating AI has been a subject of broad contention, primarily due to the common practice of training these algorithms on millions of copyrighted data points. This has raised significant concerns regarding intellectual property rights and the ethical implications of using unconsented material. Tech industry leaders have often argued that the sheer volume of data required necessitates the use of copyrighted content. However, Moonvalley aims to dismantle this argument with its ‘clean model’ philosophy.
Moonvalley’s flagship product, Marey, which is described as ‘3D-aware,’ initially debuted in a limited run in March and is now publicly accessible through a credit-based system, a common monetization model for AI video software. The company’s commitment to ethical data sourcing has garnered attention from prominent figures in the film industry, including Ed Ulbrich. Ulbrich, a renowned VFX artist and producer with credits on blockbusters like ‘Titanic,’ ‘Benjamin Button,’ and ‘Top Gun: Maverick,’ was hired by Moonvalley in June to serve as a liaison with film studios.
Ulbrich, who previously held reservations about generative AI, expressed a change of heart due to Moonvalley’s methodology. In an interview with Deadline, he stated, ‘I do think a key thing for me which was the game-changer and why I was just compelled is that at its core the idea of [Moonvalley’s] clean model of an ethically sourced, ethically trained, a legit, proper thing. No stolen pixels, no scraping of the internet. It’s done in a great way. And it is so important that it happened.’ His endorsement underscores the potential for ethically sourced AI to gain broader acceptance within creative industries.
Also Read:
- European Parliament Report Urges Sweeping Copyright Reforms for AI Data Training
- Chinese Video AI Startups Lead in Monetization, Outperforming Large Language Model Firms
This initiative by Moonvalley is not isolated. Similar efforts are emerging in other AI domains, such as the development of large language models (LLMs) trained solely on openly licensed or public domain data. These projects collectively demonstrate that powerful AI systems can be built without resorting to the controversial practice of ‘scraping’ copyrighted content from the internet. While the long-term impact and scalability of Moonvalley’s approach remain to be fully seen, it offers a compelling rebuttal to the narrative of data shortages and the perceived necessity of using proprietary material, potentially paving the way for a more ethical and legally sound future for generative AI.
