# Producer, director, screenwriter and occasional supporting actor - how we make films at SNOK > SNOK's video workshop: scenes as HTML in HyperFrames, narration and music from ElevenLabs, AI review and a human director. And how our mascot NOK came about. - Source: https://snok.ai/en/news/blog/how-we-make-films-at-snok-hyperframes-elevenlabs-nok/ - Author: Jacek Bugajski - Published: 2026-10-19 ---
The film is narrated in Polish.
Our first film starring NOK ran for nineteen seconds. It went out on 16 June 2026, and the small blue robot with an antenna has been turning up in our material on [SAP security](https://snok.ai/en/offer/sap-security/), [automation](https://snok.ai/en/offer/ai-automation/) and [master data](https://snok.ai/en/offer/master-data-management/) ever since. Nobody booked a studio for it, and there was no camera crew or editor. There was a script, a handful of AI tools and one person deciding which shot stayed and which went back for another take. That person was me, and it still is. The film above was made with the workshop it describes. ## Every scene is a web page The workshop is built around [HyperFrames](https://github.com/heygen-com/hyperframes), an open-source project from HeyGen, released under the Apache 2.0 licence, that turns code into video. Each scene is an HTML file with animation. Attributes in the markup set when an element appears, how long it stays and which track it sits on. The renderer steps through the film frame by frame in a headless browser and encodes the result to MP4 with FFmpeg. The same file always produces the same film. That matters to us for two reasons. AI agents are good at writing HTML and animation, so a first draft of a scene arrives quickly; the project's own tagline is "Write HTML. Render video. Built for agents." And edits are predictable. If I change the copy on a title card or the timing of an entrance, only that changes. With tools that generate footage from a prompt, every correction is a fresh roll of the dice and, in practice, a different scene. We have used HyperFrames since April 2026, first for title cards and end slates, now for whole motion-graphics films. [Data Factory](https://snok.ai/en/news/blog/data-factory-master-data-in-sixty-seconds/), our sixty-second film on master data, was made this way. ## Voice and music from ElevenLabs Narration and soundtrack come from ElevenLabs. We keep a catalogue of Polish voices and one rule: there is no house narrator. I choose the voice for each film, because a film about a critical SAP vulnerability and a film about a mascot call for different tones. The film in this post is read by Gregor, the same voice that narrated Data Factory. A synthetic narrator needs a script written for the ear. Give it an acronym in capitals and it may spell it out letter by letter; give it an English product name and it may improvise. One of our voices turned "Claude Code" into something close to "Plot kod". So the narration is a separate version of the script, with names spelt the way they should sound, numbers written out and short sentences. ## Four jobs on one film **Screenwriter.** I start with the point: one sentence the viewer should leave with. Scenes come afterwards, usually in the order problem, solution, evidence, close. AI helps break the idea into scenes and suggests title cards. What the film says is my call. **Director.** I approve every scene, pick the voice and set the pace. We have a readability rule: each block of on-screen text stays up for at least two seconds, or 0.4 seconds per word for longer captions. Video in the LinkedIn feed starts muted, so a caption the viewer cannot finish reading might as well not be there. **Producer.** I own formats and channels: MP4 for the website, a 4:5 cut for LinkedIn, levelled loudness. Before release a second AI agent reviews the film twice over. As producer it checks the technical side and fit with SNOK's visual identity; as viewer it judges whether the message lands and whether it would watch to the end. Its notes go back to the edit. The final say is mine. **Supporting actor.** In this film I appear in one scene, in the director's chair. My likeness was generated by AI solely from my own reference photos, and we say so. We do not generate anyone else's likeness, and client material never goes into third-party media libraries. ## Where NOK came from In June 2026 we wanted a recurring character to guide viewers through our material. The spark was [illo](https://github.com/tmchow/illo-skill), an open project by tmchow: editorial illustrations where one metaphor makes one image and a character keeps its look thanks to reference images. We did not adopt the project. We took the idea and rebuilt it on our own tools, in our own style and colours. The early versions did not survive. One was far too detailed, one came out as a shapeless blob and one looked more like an emblem than a character. I signed off the version that stayed on 14 June: a small robot with a square head, a dark visor with two eyes, an antenna tipped with a dot and a single red detail. No mouth, no superhero cape. Competent, calm, a little curious. Not a toy. The name took no brainstorming at all. NOK is simply SNOK without the S. He wears the S on his chest. ## What AI does not do for us AI shortens the road from idea to first cut. It does not decide what the film is about, which shot stays, or whether the joke at second three is funny to anyone but its author. HyperFrames is open source and anyone can sign up for ElevenLabs, so the difference comes down to the script, the editing and the person who answers for the release. At SNOK that is me, occasionally in a supporting role. *The author's likeness in the film and illustrations was generated by AI from his photographs.*