Cast
Professor Sinsky
Professor Sinsky is a university lecturer who explains world models to the students and sets their group assignment. The part needs clear, confident delivery of technical dialogue.
- Language
- Good conversational English. You do not need to be a native speaker.
- Filming
- Summer 2027
- Days
- 1 day
- Where
- Northern/Eastern Europe
- Pay
- From €175 per shooting day
- Expenses
- Travel and accommodation are covered
Audition side
The scene below has been prepared for self-tapes. You can record it on a phone and add the link to the application form below.
The lecture
Lines of other participants: Niklas (1)
Setup. Professor Sinsky is near the end of a lecture on multimodal world models for second-year media technology students. He speaks clearly and confidently, and he enjoys the material. Behind him is a big screen. The lines in italics describe what is on it, and he can play to an imagined screen. He changes the slides with a clicker and steers through a 3D reconstruction with a game controller. For more context, read the extended synopsis.
INT. LECTURE HALL — DAY
On the screen: a phone video of a university courtyard, a little shaky. A birch, a bench, a bicycle rack.
PROFESSOR SINSKY …so with video and world models, the question is no longer only: what is the next word? It becomes: what is the next frame, the next camera view, the next state of the world?
He clicks. The same courtyard, generated: glossy and cinematic.
PROFESSOR SINSKY And the representation matters. Some models work almost directly in video space. They predict frames, motion, light, texture, bodies, all together. These can look very cinematic, but they may forget the geometry of a place. A door moves. A room changes shape. A person’s position drifts.
The generated camera turns away from the bench and back. Now there are two benches. Some laughter.
PROFESSOR SINSKY Other approaches make the world more explicit. For instance, with Gaussian splatting, we can represent a captured scene as many small learned blobs in 3D space. That gives the model a kind of spatial memory. The scene is not just a sequence of pixels; it is something you can render from new viewpoints.
He steers slowly through the reconstructed courtyard with the controller.
PROFESSOR SINSKY Look at this view. This is not a photograph. Nobody ever stood here.
He drifts sideways behind the birch, where the trunk smears like wet paint.
PROFESSOR SINSKY And here, nobody should.
Laughter.
PROFESSOR SINSKY But then the obvious problem is that the world is not static. Light changes. Leaves move. Water moves. People enter the scene. So current systems often combine models: one part preserves geometry, another handles time and motion, another relights the scene, and a generative model fills in what is too complex to simulate directly. The interesting question is always: what is explicit, what is latent, and what is the model allowed to invent?
He clicks. A generated version of his teaching assistant, Niklas, walks into the empty courtyard and sits down on the bench.
PROFESSOR SINSKY So we do not ask the place to understand people. The place is one model. A person is another: body, voice, movement, each learned from additional recordings. We train the characters separately, we align them with the world, and we place them inside it.
The avatar says “The deadline is Friday.” The hall laughs.
PROFESSOR SINSKY This is not so different from what we have seen in text models. In an LLM, we tokenize, embed, attend over context, and predict what comes next. In a world model, the “tokens” might be image patches, frames, actions, camera positions, or latent scene states. The architecture changes, but the basic tension remains: memory versus prediction, structure versus hallucination, control versus flexibility.
So when you compare models, do not only ask which one looks best. Ask what it remembers, what it predicts, what it conditions on, and how it fails.
He lets that stand a moment, then checks the clock above the door.
PROFESSOR SINSKY Most of you have not handed in your project proposal, but the deadline is Friday, so please send those in now. Hand-in is two weeks out. Niklas, I’ll let you go through it.
Niklas goes through the project options. People start zipping their bags.
PROFESSOR SINSKY Time’s up. See you next… let’s see. Tuesday.
NIKLAS And you can pool—
PROFESSOR SINSKY Yes, you can pool your credits for the group project. Thanks.