The hype cycle around generative AI video has settled into a more revealing pattern. The breathless proclamations that tools like OpenAI's Sora can make filmmaking obsolete have given way to a harder reality: what Hollywood is actually building are bespoke AI models tailored to specific production challenges, not universal generators capable of conjuring finished films from typed prompts.
This distinction matters enormously. Adobe's Firefly Foundry approach exemplifies the trend with commercial-safe, IP-protected models trained for specific IP owners, reflecting a philosophy that "you don't need a model that works for everyone. You just need one trained for that use case." This is the opposite of the democratisation narrative that dominated AI discourse two years ago. Instead of levelling the playing field, studios are racing to build proprietary tools that give them competitive advantages whilst protecting their intellectual property.
AI firm Asteria has introduced Continuum Suite, described as a new-model operating system for film and television production, built on Amazon Web Services. After a script is uploaded, the system analyses characters, scenes and locations, noting props and costumes, and creates sides, call sheets and schedules, builds storyboards and animatics and makes a database for the production. This is not film generation; it is production coordination, which solves an unglamorous but genuine bottleneck that costs studios time and money.
The practical reality of current AI video models reveals why bespoke approaches are winning. What started as experimental technology producing inconsistent results has become a reliable production tool, with resolution jumping from 720p to native 4K, video length extending from 3-5 seconds to 20+ seconds, and physics simulation now producing believable real-world interactions. Yet the strongest visual diffusion models are still able to generate short clips but cannot produce longer, more coherent stories, and their outputs still seem "uncanny"—too hyper-realistic.
Amazon MGM Studios offers instructive evidence of how the industry is actually deploying this technology. Amazon MGM Studios launched a dedicated AI Studio to develop proprietary AI tools to streamline TV and film production, with a focus on areas like improving character consistency across shots and supporting pre- and post-production. One example used is Amazon's "House of David" series, which featured 350 AI-generated shots in season two. The work is embedded within existing production pipelines, enhancing what humans do rather than replacing them.
The most compelling argument for bespoke models comes from the economics of visual effects and post-production. AI is already automating cosmetic improvements, de-ageing, and dialogue replacement, with one former studio executive observing that "vanity fixes are a significant share of visual effects, and that's now pretty easy to do with AI. These tasks used to be incredibly manually intensive." This work—once requiring teams of specialists and months of labour—can now be executed with purpose-built tools trained on a studio's own footage and aesthetic preferences. That is not hyperbole; it is a genuine productivity gain.
The shift toward bespoke models also addresses the legal and ethical minefield that public tools like Sora created. Asteria is partnering with AI startup Moonvalley to build an "ethically trained" video model for studio productions, positioned as an alternative to offerings from tech giants accused of data scraping. For studios managing intellectual property and talent rights, proprietary models trained on licensed content are far less legally treacherous than relying on systems trained on scraped filmography.
The honest assessment of where AI sits within filmmaking is more measured than either boosters or sceptics suggested. AI-generated output is not yet at a quality level to drive meaningful disruption, with content that in many cases does not meet premium production standards, and there are limits to how deeply AI will disrupt end-to-end content creation and the art of storytelling in established TV and film formats. What AI can do is accelerate specific workflows, reduce labour on high-volume tasks, and lower barriers to entry for independent creators.
The filmmaker using bespoke AI tools to generate storyboards, test camera angles, or create placeholder visuals is gaining real advantage. The studio deploying custom models to de-age actors, adjust dialogue, or generate background elements is reducing costs without replacing the creative core of filmmaking. But the startup founder promising that AI will generate feature films remains trapped in fantasy. The revolution, such as it is, lies not in wholesale replacement but in granular, targeted augmentation of how films actually get made.