Can Ethically Trained AI Video Model Actually Work? This New Tool Has Your Answer

Each time I learn about yet another AI video tool, I ask myself the same questions. What was it trained on? Whom did they get permission from? And how much of it was scraped off the internet?

The reality is that ethically trained AI video models are the exception. Most use enormous databases filled with copyrighted content, such as books, movies, images, and songs, never signed away by creators to share. This has ignited all sorts of legal battles and ethics debates and made many creatives, myself included, nervous.

But today, one company claims to be doing something different.

[Related Reading: Midjourney’s New Tools: This AI Video Generation Model Is About More Than Just Animation]

Moonvalley’s Public Domain Approach to Ethically Trained AI Video Models

Moonvalley, a Los Angeles-based tiny AI firm, is pushing back on the notion that AI needs to be constructed on shaky data. Their system, Marey, is entirely trained on public domain movies. That is no pirated stuff. No scraped YouTube. Just work that legally resides in the commons.

The business began rolling out access to Marey back in March. Today, it is offered to the masses on a credit-based model, which is standard across the majority of AI video solutions.

What did catch my eye, however, was who they hired on board. Ed Ulbrich, a visual effects veteran from Titanic and Top Gun: Maverick, is now consulting with Moonvalley. Previously, he was agnostic about generative AI. But as he says, their “clean model” changed his mind.

And he is not alone. Other scientists are constructing big AI systems based on publicly available data. One new language model was trained on more than eight terabytes of open and public domain text. The team demonstrated that it is possible to construct robust instruments without using stolen content.

Sure, it is early. Moonvalley has yet to demonstrate that its training material was in the public domain. But if it stands, it would challenge the premise that big tech continues to peddle, that web scraping is the future.

I think this idea is exciting. If ethically trained AI video models can produce solid results without doing ethical shortcuts, I believe that is something to be encouraged. AI does not need to be constructed on broken trust. Perhaps now, it does not necessarily have to be.

[via Futurism; Image credits: stock photos]

---