Ben Affleck, in his role leading a film-AI company he founded in 2022, built a production-focused approach to video generative models because he judged public models - trained on peers’ films - to be commercially unusable. His 16-person shop, InterPositive, developed a workflow that starts from open video models but "unfreezes" weights and retrains only a final "cinematic" layer so outputs meet real production standards. Rather than rely on third-party datasets, the team spent eight months shooting a controlled dataset specifically for late-stage fine-tuning, then applies per-production fine-tuning on a film’s own dailies so each production retains both footage and the learned model. That combination of controlled base data plus private, per-film models is presented as the product Netflix paid $587 million in cash to acquire in March 2026.
The substance is both technical and commercial: a targeted fine-tuning technique (last-layer retraining) to align generative outputs with cinematic requirements, a data-collection strategy to avoid training on competitors’ work, and a business model that sells privacy and ownership of creative assets. Affleck’s public comments also trace a shift from skepticism about AI’s creative role toward building practical tools, while flagging broader concerns such as educational impacts and learned dependence on AI.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.