The jelly that started the jelly trend

Scott asked GPT-6 Astra for a delicious, bouncy jelly in three.js and WebGPU. The clip is short. The trend that followed is all over this site.

Sep 4, 2026last month

This is the post that kicked off the jelly trend. Scott asked GPT-6 Astra for a delicious, bouncy jelly in three.js and WebGPU, and the result is less a tech demo than a food video that happens to be a shader.

Soft bodies are a nasty test. Too stiff and it is rubber. Too loose and it never comes to rest. WebGPU is what makes the wobble cheap enough to run in a browser tab, which is why the trend could spread as copies instead of as screenshots.

Later jellies on this site — dice, cases, whatever is next — are replies to this one. Watch it as the original appetite, not as another physics toy.

The jelly that started the jelly trend

Made with

Scott

@scottstts

View the original post on X

Asked Astra to make a delicious and bouncy jelly in @threejs and webgpu https://t.co/qVqY5lpAPS

15.6K likes

likes: 15,641
replies: 179

Captured when this was curated, not live. — as of Oct 17, 2026

Keep reading

Bilawal Sidhu@bilawalsidhu4 weeks ago

Time stops being a scrub bar and becomes a shape.

Bilawal Sidhu reconstructs a geolocated 2D video as per-frame 3D, so a dog’s whole path through the shot exists in space at once — Vonnegut’s Tralfamadorians, applied to four legs.

AI 3D

Tom Krcha@tomkrchalast month

The trains are not assets. They are functions.

No 3D model files: GPT-6 Astra and Three.js generate both trains at runtime from dimensions, profiles and geometry functions — wheel motion and explode-and-reassemble included.

AI 3DGpt 6 AstraThree Js

Prasenjit@prasenx3 months ago

He never opened Blender. The render still exists.

Prasenjit had GPT-5.6 Sol, inside Cursor, set up Blender MCP and render a realistic floating MacBook. He had never opened Blender before that day.

AI 3DGpt 5.6 SolCursorBlender