Ben Affleck Reveals AI Fine-Tuning Methods for Film

Recent interview clips of actor and director Ben Affleck have gone viral, revealing his surprisingly deep technical knowledge of artificial intelligence and machine learning. Speaking at Bloomberg's Screentime 2026 conference and on GQ's "One More Question" series, the Artists Equity CEO explained how he uses Python scripts and convolutional neural networks to manipulate tensors for visual effects. Affleck detailed his hands-on approach to AI, which includes visiting OpenAI to study emerging technologies and applying these tools directly to his cinematic projects.

Affleck's technical background led to the creation of InterPositive, a 16-person AI filmmaking startup he founded in 2022. Although reports claimed Netflix acquired the startup for $587 million, Affleck clarified that the figure is inaccurate and noted he was not the sole owner. To build the company's technology ethically, Affleck raised capital and spent eight months capturing footage with specialized camera rigs. This effort produced a proprietary dataset used for the late-stage training of open-source models, linking training and inference code to execute specific post-production tasks.

For practitioners in the film industry, Affleck's methodology offers a blueprint for integrating AI without sacrificing creative control. He explained that his team fine-tunes open-source video models by unfreezing their weights and training only the final cinematic layer to meet strict production standards. This technique allows filmmakers to retain their proprietary work while generating custom, project-specific models. Affleck successfully utilized this workflow during the post-production phase of his upcoming movie, "Animals."

While many in Hollywood view generative AI as an existential threat, Affleck views the technology as an additive tool rather than a replacement for human artists. He expressed practical concerns regarding AI, pointing to a 30 percent rise in college A grades over three years and the risk of what he called "learned helplessness" in students. However, he dismissed fears of a "Skynet" scenario, emphasizing that ethical datasets and targeted model training will help the technology work collaboratively with the creative community.