ASCII jaguar — the rainforest inside the letters

2026

A jaguar made of ASCII characters runs through the rainforest of Suriname; the same jungle moves through his letters, slower. WebGL2, TypeScript, Web Audio.

From a distance you see a running jaguar; up close you see letters with the rainforest of Suriname moving through them. In the dark, loose characters flicker up and grow together into jungle, two amber eyes light up and the jaguar fills with characters from the eyes outwards. He starts walking, trots, gallops and sprints; after that he keeps galloping, with a short sprint every twenty seconds. The title JAGUAR / PANTHERA ONCA / SURINAME breaks up after a few seconds into characters that drift away with the world. Scrolling, swiping or the right arrow key gives him a burst of speed that fades in about three seconds, the space bar pauses and M turns the sound on or off. The mouse only adds a little parallax; the jaguar does not follow it.

Every frame the GPU draws four small images: the jungle outside the jaguar, the same jungle inside the jaguar (at 0.65 times the speed and brighter), the jaguar itself and the foreground. A shader turns these into ASCII, one character per cell, using the ramp ' .:-=+*#%@' from thin to full, over a dimmed version of the painted forest. Outside the jaguar the shader picks the character from the brightness of the jungle. Inside the jaguar the fur decides where characters go, so the rosettes become holes and the outline stays solid; the light of the jungle inside decides how full each character is. When the jaguar runs through a ray of light, his characters fill up.

With the real jungle in place, the jaguar at first disappeared among the characters around him. So one empty cell now runs around him, the jungle outside only gets a character above a certain brightness, in muted colours, and the foreground leaves are almost silhouettes that only darken him. His own characters have a minimum brightness and a warm amber tint. That makes him the only warm area, while the colours of the jungle stay visible in the letters.

The brief asked for a run cycle of 8 to 16 separate frames. Poses generated one by one differ in fur, rosettes and shape from image to image, and in ASCII that flickers. So the jaguar is one painted side view, cut by a script into a body and four legs and moved by a 2D skeleton with skinning. Each leg follows a foot path for walk, trot or rotary gallop: on the ground the foot moves back at exactly the ground speed, in the air it swings forward in an arc, and inverse kinematics sets the joints. A unit test checks that no foot slides over the ground. With ?debug=rig you see the silhouette with the bones, without jungle and without characters.

The forest is made of five painted layers moving at 0.15 to 1.30 times the world speed. A script of my own crossfades the two ends of each layer over 12 per cent of its width, so the layer repeats without a visible join. The near layer has no straight ground line, because painted ground slipped against the legs; the puddles are the fixed ground. Rain, mist, lianas and puddles with splashes are made of characters too.

Sound stays off until you turn it on. The rainforest, with a piha-like call and a falling tinamou whistle, and the rain are loops synthesised in advance; the volume of the rain follows the showers. The page makes the footsteps live with the Web Audio API each time a paw lands, longer and wetter in a puddle.

The code is TypeScript with Vite and WebGL2, without a framework and without three.js. On a slower device the page drops to fewer columns. Without WebGL2 it shows a still ASCII image from the real renderer, and with reduced motion on the intro is skipped and the jaguar walks. There are 129 unit tests and end-to-end tests in Chromium, Firefox and WebKit; Lighthouse scores 100 for accessibility and for best practices. A desktop loads 1.70 MB, a phone 0.98 MB. The images were made in advance with gpt-image-2 and gpt-image-1.5; the page never calls an AI service.

Related work: The jaguar of Suriname.

Technologies

Vite TypeScript WebGL2 Web Audio gpt-image-2 gpt-image-1.5

More of my work

Browse the full portfolio or read the blog.

More work