London Layer Atlas
Explore a London site through five exploded axonometric layers. Adjust colours, spacing and viewing angles, and export an editable SVG.
Build it yourself in ChatGPT or Claude
Exploded axonometric diagrams of real sites, made from OpenStreetMap data. Each demo is a small working website: turn layers on and off, change colours, spacing and viewing angle, then export an editable SVG for your boards. Every one was vibe coded from a single prompt, and you can build the same tool for your own site in ChatGPT or Claude.
Exploded map diagrams of real sites, vibe coded in ChatGPT or Claude. Open a demo and use it right here: rotate it, recolour the layers and export the SVG.
Explore a London site through five exploded axonometric layers. Adjust colours, spacing and viewing angles, and export an editable SVG.
Build it yourself in ChatGPT or Claude
Take apart a 1 km circle of Westminster as five stacked map plates. Toggle and recolour layers, change elevation and rotation, and export an editable SVG.
Build it yourself in ChatGPT or Claude
Every plate shows the same piece of the city, at the same scale and in the same position, so the layers line up exactly when they are stacked. The map data is real OpenStreetMap data, downloaded once when the demo was built and saved with the website. Opening a demo or exporting an SVG never calls ChatGPT, Claude or any other AI, and needs no account.
OpenStreetMap is made by volunteers, so a missing building or path in the data does not mean it is missing on the ground. Road widths are drawn as a simple hierarchy, not surveyed widths.
Keep the area sensible. 1 to 2 km across works well. Dense city centres produce a lot of data, so use a smaller radius there if the page feels slow.
Change one thing at a time. After the first version works, ask for one change per message: a new colour palette, a sixth layer (for example railways or contours), a different title style.
Aerial imagery has its own rules. Satellite and aerial images come from providers such as Esri with their own terms. Check that you are allowed to show and export them in your project, or use the vector base only.
The skill files help. Where a demo includes a downloadable skill, give it to ChatGPT or Claude together with the prompt to get closer to the original look.
The demos are fixed-site on purpose: the data is downloaded once, so they load instantly and cost almost nothing to host. An open platform lets any visitor choose their own site, which means fetching map data live. That changes what you need:
Start from a working fixed-site demo, then ask ChatGPT or Claude for the upgrade in plain words, for example: "Turn this into an open tool: add an address search, load OpenStreetMap data for the chosen site through a small server-side API with caching and a 2 km maximum radius, show a loading state and clear errors, and keep all styling, layers and SVG export the same. Do not call any AI model when visitors use it."
Build it in stages: address search first, then live data with caching, then limits and error messages, then deployment. Test with a small town, a dense city centre and a coastline before you share it.
Both can build these demos from the same prompt. What matters more is the mode you use: it needs to run code, download data and save files. Use whichever you already pay for, and switch if one gets stuck.
No. The prompt describes the whole website. You paste it into ChatGPT or Claude, fill in your site and check the result. Knowing how to open a folder and upload it to a host helps when you publish.
Yes. The live demos and SVG export are free for everyone. The build prompts and skill downloads are part of the membership.
Yes, with the © OpenStreetMap contributors credit (ODbL licence). Aerial imagery follows its provider's terms, so check those before you publish it.
Yes, anywhere OpenStreetMap has data. The detail varies by place, and very dense areas may need a smaller study area.
No. AI is only used once, to build the site. The finished demo is ordinary HTML, CSS and JavaScript with the map data saved inside it.
240 tested prompts, 227 worked examples and 466 before/after images for Nano Banana and ChatGPT, plus the 291-page bilingual PDF guide.