A short film that explains one AI research paper, and where it stops
One Paper Further
Part of Digital media system
A series of short films, each on one paper. Episode 1 asks how an AI-generated mountain can be found again, using the InfiniteDiffusion paper, with every spoken line traced to a place in it.
Episode 1
The paper's figures are redrawn, apart from one cropped tile credited on screen.
What it does
Episode 1 is about "InfiniteDiffusion", a paper by Alexander Goslin on generating open-world terrain, and asks how you return to the same AI-generated mountain. It follows one example from the first shot to the last, answers with the mechanism, one result from the paper and one limit, and ends on the opening picture.
How it is made
The pictures are drawn in Remotion. The voice, music and three acoustic sound effects are generated with ElevenLabs, and ffmpeg encodes the film. The diagrams are my own illustrations and are labelled as such. The only picture from the paper is one cropped tile, credited on screen, and the paper's numbers are redrawn on my own cards.
What I checked
Every spoken line sits in a claim ledger with its place in the paper. I ran no experiment of my own, did not reproduce the result and did not contact the author. Two of the paper's tables were read from its web version and are still to be checked against the PDF.
Stack
- Remotion
- ElevenLabs
- ffmpeg
- Python
How these figures are sourced
The outcome figure is agent-reported: it is taken from a test log or report written by the agents that built and ran the system. The status was read from the episode's source note and manifest. Every figure and claim on this page is tied to a source file or a CV line in the site's content.