Simulation · 25 Sept 2026 · Peter Hanlon

Piccadilly Circus

Recognisable, not photoreal.

As you know I have been playing around a lot lately in 3D visualisations and video. I find these really good story tellers, plus they are a break from CRUD semantic backends which I seem to spend most of my day job building. (At the moment at least). Given my passion is making maps, and we experience the world in 3D, the two marry together really well.

It's been a bumper week for AI model releases this week and as usual, games and 3D visualisations were some of the most popular methods of trying them out. I'm not immune to this either if you check my other posts. I haven't done any work with Blender before and many people were using this as part of their demos. It has come up a few times as I needed to paint or add texture to an existing model, usually in the form of some form of ad-creative. Anyway I decided to do a specific test of Opus 5.5 and Blender in Claude Code. I primed the original prompt with a couple of examples of my previous work, added the repos to a project so it had them to hand, and asked for a few suggestions. It came up with 4 and I set two threads of building option A and option C, later we added option E and built that. This post is about Option A and Option E.

Option A

Option A was to build a 3D city in Blender and then tell a story through the day, from morning into evening, of people moving around and different ads being displayed on a variety of DOOH screens. It sounded like a natural extension of some of my previous experiments, so we went with that. Claude built the plan, it looked good so I kicked it off. It took a while to build is all I'll say! I did check that I wasn't burning through tokens, and thankfully it was just Blender animating every frame of a 30-second 3D city walkthrough. I actually didn't get to see it until the morning as I've been burning the midnight oil this week working on our US audience proposition, so I left it running overnight and only viewed it in the morning.

Option A: a 30-second 3D city walkthrough, morning into evening, with DOOH screens changing through the day.

I was a bit underwhelmed really, not that it's terrible, it isn't, but the amount of time it took to create it, even with it being hands off, doesn't justify the quality. I shared my feedback with Claude and we re-grouped and went with a new option, option E.

Option E

The focus here was on real world accuracy. Often I find when building 3D visualisations, and other to be fair, is deciding what really matters. In this case, it's a simulation / demo. Everyone knows it, I'm not going for photo-realism. I don't want AI-fakery. I just want something that looks like a demo but is really accurate to the real world. It has to be recognisable. In my King's Cross simulation I spent a lot of time making King's Cross recognisable and accurate. That entire demo rests on that. In this case we pivoted from a cinematic sequence to an explorable 3D world, and what better place to pick than the spiritual home of UK DOOH, Piccadilly Circus.

Option E: an explorable Piccadilly Circus, built to be recognisable rather than photoreal.

I'm really pleased with the result. The accuracy is really good, good enough to be instantly recognisable. It's not perfect if you know the area well but it doesn't need to be. I paid special attention to the buses and they came out great too. The large Hero screens look great, although I think I can do more with these. I have some ideas about night time scenes where they really pop, or 3D anamorphic displays. You definitely have not seen the last of this demo!