Four recent experiments:

1. Webcam to avatar on canvas

Webcam to avatar on canvas

Using nano banana pro to generate sprites for your avatar based on one webcam shot. Thinking of Max's walky.space, Silicon Jungle's inkwell, Sprout, and Spencer's lamp experiment.

2. Object scan

Using my projector, webcam and touch frame setup to 'scan' images (take a picture of them and send to gemini and nano banana pro). Another way of jumping between the physical and digital divide. Connected to my book experiments (which is probably where I want to take this next).

3. Draw loop

From vibecheck we learned gemini did really well at organizing and commenting code within a draw loop. I thought it would be interesting to parse out the comments as section breaks, as a way to balance some form with the flexibility of having the code exposed directly (vs sliders). I'd love to do something like this with the speed/flexibility of a kind of visual repl.

4. Voxel spells

We also learned Gemini 3 is really good at voxels. I wanted something where you could spawn voxels within a world. It was interesting to figure out how to move from generating full three.js scenes to just voxels that could be inserted into the world. If we had it place each voxel directly as data it wasn't as good. What we ended up doing was having it generate code to populate a voxel array. That way it could use for loops and other procedural methods to build out the voxels but in the scene we just placed the final array.

React-three-fiber was really impressive for this. I ended up putting it down because I think you'd want to build out a full-featured 3D editor where you could move, scale and build other objects in more traditional ways. A project for another day.