Beyond pictures: Gemini helps build interactive models

Looking at a picture or trying something out

A picture of a carriage can look impressive. An interactive model can instead let you open its doors or observe its movement. Google illustrates that distinction in a collection of five projects published on 23 October 2026, which the company says were built using Gemini 4 Argon. [1]

The new development here is the collection of examples. The model itself arrived in the Gemini app on 13 October. Google describes improvements in coding and tasks involving multiple steps. These are provider claims; we have not rebuilt the projects ourselves. [1, 2]

What three of the examples show

One project turned an image of a Victorian carriage into a three-dimensional rendering with moving doors, windows and suspension that responds to bumps. Another, according to Google, combines NASA wildfire data and US Geological Survey earthquake data on an interactive globe. A third is a multiplayer browser game in which characters compete to capture flags. [1]

Their common feature is software that responds to interaction. Such an output can be explored or tried out. Our interpretation: this could help explain a subject when changing a setting makes a difference visible in the model. People still need to define what the model should teach and check that it is factually correct.

A convincing demonstration is not a validated model

Google selected these projects for its own blog. The presentation does not fully disclose how much preparation, correction or experience was involved. Some examples use additional tools; Google says the globe was built in the Antigravity coding environment. The collection therefore establishes neither a one-click process nor a typical amount of work. [1]

Our interpretation: moving parts alone do not prove accurate physics. An impressive globe likewise does not establish correct data processing or suitability as a warning system. Anyone using such a model to explain a subject should check its functions and claims individually. We have not independently checked the projects’ data or operation.

What is available in Germany

As of 24 October 2026, the Gemini web app supports Germany and German. Google lists Google AI Pro or Ultra as the plans providing Argon in the app and describes its introduction as a global rollout. Both plans are listed as available in Germany. Google says users of Gemini Apps with a Google AI subscription in the European Economic Area must be at least 18. Usage limits apply. [3, 4, 5]

This documents access to the model under the stated conditions. It does not establish that every featured project can be reproduced unchanged on every German account. Additional tools and features may have their own requirements. We have not tested these demonstrations with a German account. [1, 3]

Canvas in Gemini offers another starting point for small app prototypes. Google lists Canvas as available without a paid AI plan too; that does not mean free access to Argon. Canvas lets users request changes to apps in text, inspect a preview and view errors. Signing in is required. [5, 6]

Start with a small, checkable idea

One possible starting point would be a labelled model where selecting a component reveals an explanation. This is an editorial suggestion, not a Gemini project we have tested. A useful approach would be to describe one function, check how it responds and only then add more functions. Canvas supports iterative changes through conversation and inspection of preview errors. [6]

The interesting development is that AI can help produce an interactive prototype. A useful result still needs someone to check clear labels, correct relationships and actual usability. Google’s examples illustrate the possibility; they do not replace that work.