Simulation · 27 Sept 2026 · Peter Hanlon

Similarity Constellation

Only we have the data.

This is a follow-up post to my last post Piccadilly Circus. In testing Opus 5.5 and Blender I said I'd done 3 options and this is option C that I never spoke about. I've been working a lot on the launch of our US audience design platform, which goes live in Oct 2026 (in 6 days' time of writing this! eek), hence the focus.

This week I've been working on the vendor similarity and I already posted a video of our US Vendor Similarity module a few days ago. As part of my latest Blender test, one of the options, Option C, was to build a similarity constellation, which is similar to my atom model of similarity. Interestingly, I had tried this concept before (Christmas 2025 IIRC), or something similar. I had ideas for a Milky Way type galaxy visual where categories would form arms of the galaxy and brands would then fill out the star field.

I couldn't get anything remotely working back then, but I've had 9 months more of 3D experience and of course the AI tooling has come on significantly. I didn't do any optimisation to this model, this is what it came out as. (Actually I was really busy working out non-residential movement patterns for DOOH activation in the US and the only spare time I had was spent refining my Piccadilly Circus demo).

Option C: the similarity constellation, with no optimisation. This is what it came out as.

It's an interesting concept, and it solves the issue I had with the arms of the galaxy being really sparse. The quality of the initial video appears to be downgraded / opaque slightly so I'm sure there is more that could be done here. Of course for this to be useful I need to be able to create it on the fly in human real-time, so I think this means I'm starting to look at Blender creating objects that three.js renders in real time after fetching the data. Anyone can build the render, but only we have the data!