Gemini 3 Browser Demos: Voxel Art, Simulators and a Board Game

My team and I used Gemini 3 to create or adapt a series of browser demonstrations, from voxel art to a custom board game. Several are playful visual prototypes, and some needed iteration or had visible flaws. The video includes a sponsored segment for Zapier.

We started with a voxel art generator supplied by the Gemini team and continued the conversation to make it generate procedural robots. After a few back-and-forth turns, the demo could create a robot, break it into pieces and reassemble a new one. A custom request for a retro instant camera produced something recognizable, although I did not think the colors were quite right.

The interface lets us import and export a JSON blueprint, request a custom object, generate a random robot and pause or resume the spinning animation. In this version, generating a new robot kept the previous colors; refreshing the page produced a different palette. The shattering and reassembly effects are particularly satisfying to watch.

Next we prompted an application for a mirrored room with metallic textures and a voxel character. In the resulting browser demo, reflections change as the camera moves and metallic objects float through the room. There are visible artifacts and the rendering is not perfect, but I like the way reflections appear on multiple surfaces. The video does not independently establish the precise rendering technique or performance.

We also asked Gemini to research financial bubbles and build an interactive storyboard about the AI boom. It compares historical examples, presents a Minsky-cycle framework and includes a game in which actions such as building infrastructure or raising money change the displayed risk. In my demonstration, repeated fundraising eventually pops the simulated bubble.

The page displays figures about capital expenditure, revenue and energy use, then labels the AI boom a productive bubble in its generated assessment. Those numbers and the verdict are outputs of the demo, not independently verified financial analysis or a forecast established by the video. The simulator illustrates an idea rather than proving how markets will behave.

A sponsored segment introduces Zapier's Unified Copilot, which the sponsor presents as a conversational way to build automations across its products. The video also describes integrations and a human review step. Those are sponsor claims and 2025 product details, not a test of every workflow or current availability.

Another voxel tool takes an image and creates three-dimensional assets. I show examples including a cat with a jetpack and a tree on a floating island, then upload a well-known Muhammad Ali photograph. The prompt asks for three distinct voxel representations, and the resulting figures are fairly simple but recognizable; the interface offers downloads.

For a particle-collider idea, we asked for an engaging simulation and also suggested a tokamak-like plasma demonstration. The resulting app offers two views. It runs slowly in my browser, and the AI analytics panel shows an unstable connection after the API quota was exhausted. The visuals alone do not verify that the underlying physics is accurate.

A powers-of-ten animator uses generated images to move between scales. In one sequence it travels from the solar system toward a cell and smaller scales; another moves from a satellite view toward a street and building. The transitions are fun, but I notice odd elements and imperfect geography, so the generated imagery should not be treated as a scientific or geographic visualization.

We prompted an interactive N-body gravity sandbox with controls for mass, position and initial velocity. I add bodies, watch a collision and introduce a passing star or a black hole. The displayed paths react to the inputs, but the video does not validate the numerical method or establish that the black-hole behavior is physically realistic.

The golf-swing analyzer takes an uploaded video and presents a score and suggestions. In this example it gives the swing 78 out of 100, highlights several movement measures and compares the motion with professional golfers. The clip demonstrates generated feedback, not a validated measurement or a substitute for qualified coaching; I do not independently verify the score or comparisons.

A hand-tracking visualizer responds to movements from one or both hands and offers particle styles such as cosmic energy and glacial shards. The interaction makes a playful browser effect; the video does not test its tracking accuracy or demonstrate a validated fluid-dynamics model.

For an Earth visualization we asked for a rotatable globe, clouds, adjustable conditions and a timeline. The interface shows controls for carbon dioxide, temperature and cloud layers, along with weather and satellite views. These are interface features observed in the demo, not evidence that the weather, climate or satellite calculations are scientifically accurate.

The dot-matrix converter turns a photo I took around the Bay Area into particles. I try effects that pull or scatter the dots, then reconstruct the image. The visuals run in the browser and can be replayed with different styles.

Finally, we created a Monopoly-style board-game generator that accepts a theme. I show a Donkey Kong-themed board with properties, dice, pieces and simulated opponents that can trade or negotiate. The demonstration shows a few turns and is not perfect; it does not establish that all game rules, negotiations or generated content work reliably. I think continued iteration could make it more compelling.