The Technical Context
I looked at this case without the romanticism: the fact of a 90-minute runtime is impressive, but my immediate interest isn't the "wow" factor, but how it was assembled. Frankly, today's AI implementation in video almost never means one prompt and a finished feature-length result.
I would bet on a familiar scheme: a script outline, a batch of short scenes, character references, followed by editing, sound design, cuts, and manual rhythm adjustments. This is exactly how long AI videos are currently made, even if announcements frame it as "we generated a film."
As of May 2026, mass-market text-to-video models have no confirmed mode where I can press a button and get a cohesive 20-90 minute episode with stable characters, physics, and dialogue. Sora 2, Veo 3.1, Runway Gen-4.5, Kling 3.0, and similar systems have gotten much better at clips, scene extension, and consistency, but it's still clip-based production, not an autonomous director.
Here’s where I see real progress: temporal consistency has improved noticeably, camera movement is less “floaty,” faces break less often between frames, and the multi-shot workflow finally doesn't look like research hell. Plus, some models are already quite helpful with audio and lip-sync, which significantly cuts down on manual assembly.
So, the news isn't that someone conquered cinema with a single prompt. The news is that the AI filmmaking pipeline has matured enough for long-form content, provided the team has enough discipline and patience.
Impact on Business and Automation
For studios, marketing, and edtech, this is a positive signal: long-form content can now be assembled faster and cheaper than a year ago. It’s not free and not without people, but the barrier to entry has dropped.
Who wins? Teams that know how to build an AI architecture around scene generation: script, shot list, character control, editing, voice-over, QA. Who loses? Those still waiting for a “magic button,” not understanding that automation with AI works like an assembly line, not a magic trick.
I see this with my clients: the most valuable asset isn't the model itself, but the combination of tools and rules that yields a repeatable result. At Nahornyi AI Lab, we solve these exact bottlenecks when the goal isn't just to play with neural networks, but to build a working pipeline for content, training, or media production.
If your team is already drowning in manual video assembly, let's break down the process and eliminate the chaos. At Nahornyi AI Lab, I can help you build AI automation for your production so that long-form content stops being an experiment and starts meeting deadlines and budgets.