The lab
Web experiments.
A few of these came out of client work. The rest we built to find out how something worked. If something here would be useful on your project, it can be.
Voice clones live
Six voices you'll recognise, none of them real - CosyVoice given a few seconds of each and asked to read the same line. Here to be judged, not used.
Have a listen →
MiniMax H3 workflow live
A shot blocked out in Blender, then rendered as finished footage by MiniMax H3. Sweep between the previs and the final cut.
See the comparison →
Image to mesh live
One picture in, one 3D mesh out, using Hunyuan3D 2.1 on the machine under the desk. Two goes at it: a character that works, and a castle that doesn't.
Have a look →Reel live
Ten short clips made with AI, played one after another and back to the start. Nothing here was filmed.
Watch the reel →
Signal Lab live
Three.js audio visualisers driven by one Web Audio analyser. Switch style without a reload, tune everything live, and share an exact configuration by URL.
Open the lab →
Web-GL live
Drop in a logo or an image, pick an animated shader, tune it, and copy the snippet straight out. The hero on this page is its Smoke Veil preset.
Make something →
Elk live
A zero-dependency diagram renderer that speaks Mermaid syntax and outputs raw SVG. Twenty-six diagram types, custom parsers, no Mermaid underneath.
Open the renderer →More on the way
Scroll-driven transitions, video manipulation in the browser, live capture tools and a few things we're not ready to show yet.
Ask us what's next