Natasha Lyonne to direct sci-fi film powered by Moonvalley's AI

Hollywood has spent the last couple of years treating generative AI like a radioactive topic in public, even while quietly exploring it behind closed doors. Now, a prominent indie darling is stepping directly into the blast radius.

Poker Face star and filmmaker Natasha Lyonne has announced she will direct and star in a new sci-fi feature titled Uncanny Valley. The project will heavily integrate elements generated by AI models to build out its fantastical world.

A prestige indie crew meets clean data

The creative lineup is surprisingly heavyweight. Lyonne co-wrote the script with The OA creator Brit Marling, who will also co-star in the film. The story centers on a teenage girl whose life is upended by a wildly popular virtual reality game. Lyonne teased the vibe as a Wachowski-style trip, comparing it to Dianne Wiest and Diane Keaton taking a fast-talking stroll through The Matrix.

Virtual reality pioneer Jaron Lanier is on board to produce, while Asteria, an AI-driven production studio co-founded by Lyonne and producer Bryn Mooser, is handling the visual generation.

To dodge the inevitable industry backlash, Asteria is leaning hard on its tooling choice. They are using Marey, a text-to-video AI model built by startup Moonvalley. The company pitches Marey as one of the first truly clean models on the market, claiming it was trained entirely on licensed material with original creators properly compensated.

Production dates have not been locked down yet, and the team hasn't announced whether it will head to theaters or streaming.

Why it matters

This project feels like the ultimate test balloon for generative AI in mainstream entertainment. Up to this point, generative video has mostly been the domain of tech demos, viral social clips, and experimental indie shorts. Having credible creators like Lyonne, Marling, and Lanier attach their names immediately shifts the debate from tech hype to serious cinematic application.

From an industry standpoint, Asteria is playing its cards smartly. Just last month, over 400 artists signed an open letter slamming OpenAI and Google for scraping copyrighted work, and memories of the 2023 strikes over digital likenesses are still fresh. By anchoring the workflow to Marey and its licensed training pool, Asteria gives itself a defensive shield against copyright lawsuits and labor protests.

However, we think the real hurdle lies ahead in execution. Clean models often face severe dataset constraints compared to scraped mega-models, raising doubts about whether the fidelity can genuinely sustain a full-length feature. More importantly, audiences do not buy movie tickets to celebrate ethical compute pipelines. Solving the copyright headache is just table stakes; proving that an AI-assisted pipeline can produce a genuinely compelling narrative film is the real test.