Ben Affleck went from movie star to subject of a technical discussion after clips from recent interviews spread in which he explained concepts such as machine learning, neural networks, transformers, tensors, graphics processing units, and inference. His comments were not limited to terminology; he directly connected these concepts to visual-effects workflows and filmmaking.
During his appearance on GQ magazine’s One More Question series, Affleck said he had been interested in computers since childhood and that the transition in film production from analog film to digital prompted him to delve deeper into the technical side. He also said he could write simple Python scripts and that machine-learning technologies had been used in visual effects for years.
From Image Analysis to Visual Effects
Affleck explained that convolutional neural networks had been used to detect patterns and features, such as identifying the edges of elements within an image, helping separate a green-screen background and replace it with other elements. He also described a tensor as a digital representation of an image containing information such as the batch number, frame, and color values for each pixel.
The simplicity of his presentation of complex concepts helped the clips spread, particularly because he defined the terms for the interviewer as he used them. He also noted that he had benefited from his position in Hollywood to visit organizations working on artificial-intelligence technologies, including OpenAI, before deciding to participate directly in the field.
Custom Data Instead of Relying Entirely on General Models
Affleck said his artificial-intelligence startup for filmmaking sought to develop an approach that integrated the technology with artists while taking into account issues involving image, identity, and the associated rights. For this purpose, he said he raised funding and filmed over approximately eight months using a large number of cameras and pieces of equipment to create a proprietary dataset.
According to his explanation, the data was designed to train open models at advanced stages to perform specific tasks, linking the training code to the inference code in support of practical production uses. The company was later sold to Netflix earlier in the year. Reports circulated a figure of $587 million for the deal, but Affleck said the figure was incorrect, explaining that he did not own the company in full.
What Changes in Practice?
In another conversation with Bloomberg’s Lucas Shaw during the Screentime 2026 conference in Los Angeles, Affleck described an approach based on taking open-source models, modifying their weights, and then training them according to specific cinematic criteria. According to his explanation, a filmmaker can retain their own work and benefit from the knowledge resulting from training a model customized for the film or task they are working on.
Affleck used this approach in his film Animals, where artificial intelligence helped during post-production. The significance of this proposition is that it does not present artificial intelligence as a comprehensive replacement for filmmaking, but as a specialized tool trained on data and contexts determined by the production team. However, the article does not clarify details about the model used or the nature of the data and the rights to use it—points that remain essential to evaluating this approach in practice.
As for the risks, Affleck said he was not concerned about scenarios resembling “Skynet,” but rather about artificial intelligence’s impact on education, responsible use, and what he described as learned helplessness. He also believed that the technology would not take over filmmaking in a significant way, but would instead be an addition to it.