The jaguar of Suriname — a 60-second nature film

2026

A 60-second nature film in the browser about the jaguar in Suriname: painted images in WebGL, Dutch narration and sound driven by Web Audio.

The jaguar of Suriname is a nature film of exactly 60 seconds that is not a video: the browser plays painted images, text and sound as layers on a single timeline. There are eight scenes: the rainforest, the encounter, a map of Suriname, the jaguar on a fallen tree, the jaguar in the river, the animals it hunts, the threats and a closing shot in the sunlight. Everything loads before 'Start de reis', with a progress bar, so the film never has to wait halfway. Space or K pauses, M turns the sound on or off, C does the same for the subtitles and R starts again. After the film, 'Meer ontdekken' offers more about tracks, food, range and protection.

The Dutch narration uses 116 words. It calls the jaguar the largest cat of the Americas, says that more than ninety per cent of Suriname is covered in forest and that the jaguar is an excellent swimmer that hunts capybaras, peccaries and tapirs. The map shows the Central Suriname Nature Reserve, 1.6 million hectares in size and World Heritage since 2000. Every claim is listed with its source in a fact sheet, checked with the IUCN, Panthera and UNESCO, among others. Whatever was not solid enough was left out. How many jaguars live in Suriname is unknown, so the film gives no number. And because 'the strongest bite of all cats' is disputed, the narration says 'one of the strongest bites'.

The images were painted in advance with OpenAI's gpt-image-2, in the style of a hand-painted animated film: gouache and watercolour on paper. The same model sheet was sent with every jaguar image, so it is always the same animal. The plan was to make five scenes as video with Sora 2, but OpenAI had shut that model down on 24 September 2026. So each jaguar scene has one painting with two or three variants in which only the jaguar moves, with 0.7-second dissolves, like hand-drawn animation with few drawings. The animals and the foreground leaves are separate cut-outs made with gpt-image-1.5.

A custom WebGL shader brings the paintings to life: water ripples and glints, the air shimmers and the jaguar breathes. Where it breathes comes from a mask that a script computes by comparing each pose with the same painting without the jaguar and keeping the largest connected area. On a phone that mask also decides the crop: the jaguar stays whole in view, over a blurred copy of the painting. Move the mouse close to its head while it looks at you and it blinks.

The voice was made with gpt-4o-mini-tts, voice marin, as eight separate sentences that a script places at their time. The first version ran into the next sentence in three places, because the voice speaks about 1.8 words a second, not the 2.3 that had been assumed. Three sentences were shortened, and 'voor ruim negen tiende' became 'ruim negentig procent', because spoken it sounded like 'negentiende' (nineteenth). Each clip was then turned back into text with gpt-4o-transcribe; all eight came back word for word. Music with pads and kalimba, forest sounds and effects were synthesised with a script of my own, because OpenAI has no model for music or natural sound.

The film is built with SvelteKit 2 and Svelte 5 as a static site, without GSAP, three.js, tracking or cookies. The Web Audio API clock is the timeline: four sound layers start on the same sample, and pausing stops that clock, so picture, subtitles and sound cannot drift apart. The map uses the borders and rivers from Natural Earth, simplified to 35 kB. The website calls no AI service; an end-to-end test fails as soon as the page requests anything outside its own site. There are 19 unit tests and 8 end-to-end tests, which run in Chromium, Firefox and WebKit. The start screen weighs 0.7 MB, the whole film 7.3 MB on a desktop and 5.4 MB on a phone. With reduced motion on, the camera stands still and no particles float.

Related work: ASCII jaguar, Suriname.

Technologies

SvelteKit WebGL Web Audio API TypeScript OpenAI gpt-image-2 OpenAI TTS

More of my work

Browse the full portfolio or read the blog.

More work