APIMODELSAPIMODELS

20+ Gemini 4 Carbon Games Prompts

20+ prompts. 223 things people built, tested and compared with Gemini 4 Carbon in its first weeks: 3D scenes, games, simulations, films made from code, real bug fixes and research, and dozens of same-prompt tests against Claude Fable 5.2 and Gemini 4 Argon. Each is credited and linked to the original post, with the prompt where the creator published one.

← All 223 Gemini 4 Carbon prompts

Last updated: · Gemini 4 Carbon API — pricing & playground · API docs

TL;DR

Gemini 4 Carbon (model id gemini-4-carbon) is Google's deeper-reasoning Gemini 4 model. Google built it on Gemini 4 Argon and describes the difference as more thinking tokens per turn; it opened to vetted cybersecurity defenders on 28 October 2026 and to every paid Gemini API developer on 10 November. It reads text, images, video, audio and PDFs with a 1,048,576-token context and writes text, so the scenes, games and films here are code it wrote (Three.js, WebGL, Blender Python, CUDA C++, Web Audio), mostly run by Google's Antigravity agent. Of the 223 cases, 18 include the prompt as published, 33 only the creator's request and 172 no prompt; we never fill one in. In same-prompt tests creators usually found Claude Fable 5.2's results more polished and Carbon's cost a small fraction of Fable's. On apimodels.app Carbon costs $1.10 per 1M input tokens and $5.50 per 1M output through 30 November 2026 and $2.20 / $11 from 1 December, against Google's $2 / $10 and $4 / $20.

Source: Google · Gemini 4 Carbon (the model vendor’s official site)

Hover to preview · click for sound

Chrome Dino upgraded by four rounds of self-review

Prompt not published

No prompt published. The author had Carbon judge its own screenshots like a VFX supervisor and fix whatever looked fake, four times. Over 4 passes (about 2 hours, about $7.4) the dino went from a voxel block to a striped raptor, the white void became golden hour, rocks became dead trees and cacti, and paper birds became flapping pterodactyls.

Clip:
Video, 40 s. About 13K views and 116 likes on X as of 12 November 2026.
Numbers the author reported:
4 passes· ~2h · ≈ $7.4
Note:
Posted 30 Oct, before the 10 Nov API launch. The author says the prompt is in the replies. Builds on the same author's Chrome Dino comparison (2115953941808550313); test run through a third-party API.
video by @noclipepeSource
Hover to preview · click for sound

Cozy puzzle game built entirely in AI Studio

Prompt not published

No prompt published. A cozy puzzle game built entirely with Carbon in Google AI Studio and published at a playable link; a full playthrough takes about 10 minutes.

Clip:
Video, null s. About 46K views and 90 likes on X as of 12 November 2026.
Tool:
AI Studio
Numbers the author reported:
about 10 mins for full playthrough
Note:
Playable at the-quiet-elevation.ai.studio (link in the post).
video by @chanelluuhSource
Hover to preview · click for sound

Playable V12 hypercar coastal racing game in one HTML file

Prompt not published

No prompt published. A playable 3D V12 hypercar racing game along a coast, in a single HTML file with Three.js. It includes a 9,250 RPM engine sound, dual-clutch shifts, tunnel reverb, hairpin drifts and 340 km/h viaduct overtakes at 60fps.

Clip:
Video, 32 s. About 1,871 views and 59 likes on X as of 12 November 2026.
video by @StatsWireSource
Hover to preview · click for sound

Seaplane game with an infinite ocean and random islands

Prompt not published

No prompt published. Given a 3D seaplane prompt, Carbon built an infinite ocean with randomly spawned islands and full controls.

Clip:
Video, 50 s. About 54K views and 56 likes on X as of 12 November 2026.
Note:
Posted 6 Nov, before the 10 Nov API launch. Written in the third person; the post does not say who ran the prompt.
Hover to preview · click for sound

Open-world pelican bike game across Google's campus

Prompt not published

No prompt published. The author took the pelican-riding-a-bicycle benchmark into a playable game using Carbon and /swarm in Antigravity. Multi-agent critique loops ran in 5 waves: subagents generated Nano Banana and Omni references, rigged assets in headless Blender, wired Three.js bike physics, wrote custom WebGL2 shaders and added procedural Web Audio effects with a Lyria track.

Clip:
Video, 77 s. About 1,249 views and 55 likes on X as of 12 November 2026.
Tool:
Antigravity
Numbers the author reported:
5 waves
Note:
The author calls it "our Mountain View campus"; the video is the one in Sundar Pichai's launch post. Reposted by many accounts (@choblin29, @alexmichael, @EnvolDev, @masahirochaen, @Saboo_Shubham_, @songguoxiansen); this is the post where the maker describes it.
video by @agenticjaySource
Hover to preview · click for sound

Glassy Jenga physics game you can play by sound

Prompt not published

No prompt published. A physics game called Glassy Jenga with shattering glass blocks. You can play by tapping the blocks for sound, or switch to a sandbox and smash things; it is live at a public link.

Clip:
Video, 57 s. About 18K views and 42 likes on X as of 12 November 2026.
Note:
Playable at vitrail.andrewvoirol.com (link in the post).
video by @AI_AndrewSource
Hover to preview · click for sound

Rebuilding a Fable-made shooter from its gameplay video

Summary

Only the creator's request is public, not a full prompt: "能不能做一个一模一样,或者正面超越它的东西?"

What it is:
The author fed Carbon in Antigravity the gameplay video of Saltline, a 3D web shooter Fable 5.2 built in an 8-hour, $550 run, and asked for an identical or better game. After 48 minutes and $3.82 it produced a 638.5KB single HTML file with Three.js inlined plus new mechanics; a visual self-check loop with headless Chrome screenshots caught and fixed three bugs, such as a barrel rotated 22.5 degrees.
Clip:
Video, 52 s. About 7,165 views and 33 likes on X as of 12 November 2026.
Tool:
Antigravity
Numbers the author reported:
48 分钟后 · 3.82 美元 · 7.64 美元 · 638.5KB · 103KB · 39 轮调用(562.5 万总输入 / 468.4 万命中缓存 / 14.7 万输出 token) · 8 小时、烧掉 550 美元
Note:
Saltline was made with Fable 5.2 by Anthropic engineer Felix Rieseberg. Not like-for-like: that run started from a three-line prompt, while Carbon got the finished game's video. $7.64 is the author's estimate at the post-promo price.
video by @mylifccSource
Hover to preview · click for sound

Boat racing game from a single prompt

Prompt not published

No prompt published. The author says one prompt was all it took to get this game from Carbon.

Clip:
Video, 31 s. About 1,176 views and 27 likes on X as of 12 November 2026.
Numbers the author reported:
just 1 prompt
Note:
Subject (a water racing game with a lap counter, minimap and speedometer) is from the video thumbnail.
Hover to preview · click for sound

Pelican riding a bicycle turned into a game

Prompt not published

No prompt published. The author turned the pelican-riding-a-bicycle test into a game with Carbon.

Clip:
Video, 15 s. About 1,879 views and 19 likes on X as of 12 November 2026.
video by @xluchuanSource
Hover to preview · click for sound

Game rebuild from an Omni video that became a new game

Prompt not published

No prompt published. The source was a video made with Omni, so the author expected Carbon to recreate it even better than Fable did. Instead Carbon built a completely different game, which is playable at a public link.

Clip:
Video, 50 s. About 779 views and 14 likes on X as of 12 November 2026.
Compared with:
Claude Fable 5.2. Instead of recreating the video like Fable, Carbon built a completely different game.
Note:
Playable at gigabitgamelab.com (link in the post).
Hover to preview · click for sound

One-shot Mario-style 3D platformer via Vertex AI

Prompt not published

No prompt published. One shot with no external tools, through Vertex AI: a playable Mario-style 3D platformer with multi-joint skeletal animation, 4 levels, moving-platform physics, Web Audio music and 10 sound effects. It came to 2,599 lines (99.6 KB) in one HTML file, 73,907 tokens in total and 558.7 seconds end to end, at a reported bill of $0.45.

Clip:
Video, 37 s. About 4,366 views and 13 likes on X as of 12 November 2026.
Tool:
Vertex AI
Numbers the author reported:
2,599 行 / 99.6 KB · 38,289 Tokens · 35,618 Tokens · 73,907 Tokens · 558.7 秒(9 分 18 秒) · $0.45
Note:
Billed at the promo rates valid until 30 Nov. @KOF97AI later posted a near-identical write-up with the same figures but said it ran in Antigravity. The quoted earlier post by this author compares Carbon and Argon on a one-line pelican SVG prompt via Vertex AI.
video by @Delroy715Source
Hover to preview · click for sound

Simple racing game from a first test

Prompt not published

No prompt published. The author's first Carbon test: a simple racing game that the author found very polished.

Clip:
Video, 68 s. About 2,087 views and 12 likes on X as of 12 November 2026.
Hover to preview · click for sound

Desk Racer, a miniature kart game in the browser

Summary

Only the creator's request is public, not a full prompt: "build a 3D racing game"

What it is:
Asked to build a 3D racing game, Carbon made Desk Racer, a miniature kart racer that runs in the browser: 4 game modes, Rapier3D physics, Three.js graphics, multiple camera angles, dynamic engine sounds and 60 FPS play.
Clip:
Video, 76 s. About 1,022 views and 9 likes on X as of 12 November 2026.
video by @anuragdotdevSource

NIGHTSHIFT highway racing game built twice: Carbon vs Fable 5.2

Prompt not published

No prompt published. From one long prompt, Carbon (High) in Google Antigravity chose the stack, wrote a rainy midnight highway racing game (WebGL2, three.js, Rapier), modelled the player car in Blender with a Python script, pulled in CC0 assets and tested it in a browser, with no human-written code. Fable 5.2 built the same game in Claude Code at xhigh as a lead agent delegating to Opus 5.5 subagents.

Where:
Blog post, 2026-11-11
Tool:
Google Antigravity (the write-up says the Antigravity CLI)
Numbers the author reported:
Carbon: about 2 h 8 min · Carbon: about $10 (estimated at third-party per-token prices) · Carbon: 20 source files / 7,705 lines · Car model: 89,900 triangles · Fable: about 16 h 36 min · Fable: "$300+" ($236 logged) · Fable: 119 files / 35,831 lines
Compared with:
Fable 5.2. No single winner: Carbon costs far less and built a decent game, while Fable treated it as a full project with planning, parallel subagents and a self-playing test driver.
Note:
Described as one long prompt, not published. Carbon's cost is the author's estimate at third-party per-token prices.
Also:
https://github.com/EddizEge/nightshift
Watch on the original post

Cozy medieval diorama builder in a single 930 KB HTML file

/swarm 请你做一个类似 tiny glade 游戏的 AAA 级别的游戏,画质要用上所有最新的手段,能在浏览器环境运行的, html 版本

What it is:
A /swarm multi-agent run in Antigravity built a browser diorama builder with procedural stone walls, auto-carved arches and bridges, 195k-blade GPU grass and procedural WebAudio, with every asset generated in code. A making-of report is linked from the README.
Where:
GitHub, 2026-11-11
Tool:
/swarm (Antigravity's multi-agent mode, per the prompt)
Numbers the author reported:
1 h 41 min swarm run · 9 agents · 22,910 LOC with 0 merge conflicts · 4.14x peak parallel speedup · 4-round visual/shader QA loop · final file 929.6 KB · 195k-blade GPU grass
Note:
Clone of an existing commercial game's concept and name (Tiny Glade).
Watch on the original post
by FireTableSource
First-person battle mage game, judged blind: Fable 5.2 vs CarbonYouTube

First-person battle mage game, judged blind: Fable 5.2 vs Carbon

build a first-person battle mage game. Sword in one hand, spells in the other, enemies worth fighting, a way to win, a way to lose, and it has to start from one double-click.

What it is:
Each model, on its highest setting in an empty folder, got one shot and one hour with no follow-ups to build a first-person battle mage game. The folders were relabelled GAME A/B and judged blind before the key was revealed.
Where:
YouTube video, 2026-11-11
Numbers the author reported:
one shot, one hour each, no follow-ups
Compared with:
Fable 5.2. The verdict is only in the video.
Note:
The title says one model won, but the winner is only revealed in the video. Harness not named.
3D kart racer on a giant dinner table, $7 per modelYouTube

3D kart racer on a giant dinner table, $7 per model

Prompt not published

No prompt published. Fable 5.2 and Carbon each got $7 for the same job: a full 3D kart racing game on a giant dinner table, with phone controls, a computer racing mode and three difficulty levels the models had to design.

Where:
YouTube video, 2026-11-11
Numbers the author reported:
$7 cap per model · three difficulty levels
Compared with:
Fable 5.2. The title says Fable 5.2 was beaten by Carbon.
Note:
Details and scoring are only in the video.
哔哩哔哩

Browser-playable FPS game from a one-sentence prompt

Prompt not published

No prompt published. Carbon built a web-playable FPS game from a one-sentence prompt.

Where:
Bilibili video, 2026-11-11
Note:
No numbers in the description; the prompt is not published.
video by SmartDeng_Source
哔哩哔哩

Bridge-building contest re-run with Gemini models in Antigravity

Prompt not published

No prompt published. The author re-ran another creator's "AI algorithm contest, ep. 9: bridge building" with Gemini models in Antigravity. Carbon did badly for unknown reasons, while a new gray-test model did well and fast.

Where:
Bilibili video, 2026-11-11
Tool:
Antigravity
Compared with:
new gray-test model. Carbon did badly for unknown reasons; the new gray-test model did well and fast.
Note:
The author blames Antigravity as a drag on the results.
video by tzt诶喂Source
Fall Guys-style party game from scratch: Carbon vs Argon vs Fable 5.2YouTube

Fall Guys-style party game from scratch: Carbon vs Argon vs Fable 5.2

Prompt not published

No prompt published. The same prompt went to Carbon, Argon and Fable 5.2 to build a Fall Guys-style party game from scratch, with no human coding and at most 3 bug fixes.

Where:
YouTube video, 2026-11-10
Numbers the author reported:
at most 3 bug fixes
Compared with:
Argon, Fable 5.2. The verdict is only in the video.
Note:
Results only in the video; the description gives no numbers, prompt or verdict.
video by GameAGISource

How to write great Gemini 4 Carbon prompts

  1. 1Write a spec, not a wish. Creators who gave Carbon one vague line mostly got average results; one who ran 15 projects summed it up as: don't make wishes at it. The striking results came from briefs that named the stack, the deliverable and what done looks like, up to a spec of about 10,000 characters for a three-minute violin concert film.
  2. 2Show it a reference. Carbon reads video. One creator gave it the gameplay video of a shooter Claude Fable 5.2 had built and asked for the same game or better; another gave it Anthropic's own Fable 5.2 demo video to rebuild as a web page.
  3. 3Make it look at its own output. The shooter rebuild ran a headless-Chrome screenshot loop that caught and fixed three bugs on its own, in 48 minutes for $3.82. Another creator had Carbon review its screenshots like a VFX supervisor four times, about two hours and $7.40, turning a block of voxels into a convincing dinosaur.
  4. 4Name the format. "One HTML file. No external assets." is half of the 5,000-drone light show prompt. Simon Willison had to add "build that into a standalone HTML file" to get a page he could open on its own.
  5. 5Split big builds across agents. The largest builds used Antigravity's multi-agent /swarm mode: nine agents and 22,910 lines for a diorama builder in 1 hour 41 minutes, and five waves of agent critique for an open-world game.
  6. 6Follow up. One creator posted the same island and castle after the first prompt and after three follow-ups; the fire went from nine orange cubes to something close to real. The gap, in the author's words, came from the follow-ups.
  7. 7Check what it says it finished. In one six-app test every one of Carbon's final summaries claimed at least one feature that turned out broken or overstated. Ask it to click through the result in a browser and report what it saw.

Prompt formula

Carbon build brief = the deliverable and format (one HTML file, an MP4, a repo) + the stack and constraints + what done looks like + a reference (an image or video it should match) + a self-check (open it in a browser, screenshot, fix) + a stop condition or time and token cap

Example (our template, not a creator's prompt): Build a browser game: a night-time arcade racer on a neon highway. Three.js from a CDN, music and sound in Web Audio, everything else in one index.html. Match the mood of the attached video. Done means: it loads with no console errors, steering and crashes work, and a lap takes about 90 seconds. Open it in a headless browser, take screenshots at the start and mid-lap, fix what looks wrong, then tell me what you checked. Do not ask questions.

Frequently asked questions

Can Gemini 4 Carbon make 3D scenes, games or videos by itself?

Not directly. Carbon takes text, images, video, audio and PDFs and returns text. Every scene, game and film here was rendered from code it wrote: Three.js and WebGL pages, Blender Python, CUDA or C++ programs, Web Audio for sound. In a chat window you get the code or an HTML file; an agent such as Antigravity, Cursor or GitHub Copilot can run, render and test it.

Which tool did people use with Carbon?

Mostly Google's Antigravity, often with its /swarm multi-agent mode. Others used Cursor, GitHub Copilot, AI Studio, Vertex AI or the API from their own harness. Where the creator said, the card names the tool.

Is Carbon better than Claude Fable 5.2 for this kind of work?

Usually not on polish, often yes on cost. In most same-prompt tests here Fable 5.2 produced the better-looking, more complete result, while Carbon cost a fraction as much and sometimes won outright: $7.54 against $41.71 for three jobs in one test, where Carbon won two of the three.

Is Carbon better than Gemini 4 Argon?

Mostly yes, at the cost of time and tokens. The two cost the same per token; in these tests Carbon usually produced the stronger result and took several times longer, for example 81 minutes against 14 in one Blender test that Argon had failed. Google still recommends Argon for most everyday coding.

How much did these builds cost?

As the creators reported: $0.45 for a one-shot 3D platformer of 2,599 lines, $3.82 for the shooter rebuild, about $7.40 for four rounds of self-review and $27.68 for three long builds. Cost depends on the thinking level and how long the agent runs; thinking tokens are billed as output.

How do I run Gemini 4 Carbon?

On apimodels.app, call model gemini-4-carbon on the OpenAI-compatible /v1/chat/completions endpoint or the native Gemini endpoint with one key, from anywhere including mainland China. Raise the output limit to at least 32,768 tokens, stream long answers and allow a 15-minute timeout, because Carbon thinks for a while before it writes.

Can I reuse these prompts and videos?

The works belong to their creators. Clips from X play from X's own servers, videos on YouTube and Bilibili play in their own players, and every card links to the original; we do not host or re-upload any of them. Published prompts are quoted with credit, in full when short and excerpted when long. If you made one of these and want the card removed or the credit corrected, email us.

Run these prompts on apimodels.app

Gemini 4 Carbon is live here alongside 160+ other image, video, audio and language models — one API, one key, pay as you go, with $0.10 of trial credit on sign-up.

This library is open source on GitHub

The whole collection is on GitHub as Markdown and JSON: all 223 cases with credits and links to the original posts and pages, the tools and numbers each creator reported, and a summary of the same-prompt tests. The videos and images stay with their creators and are linked, not re-hosted. Prompts up to 400 characters are quoted in full; longer ones are excerpted, and unpublished ones are left out. Our own write-ups are licensed CC BY 4.0. Open an issue to add a case or correct a credit.

View on GitHub

Other prompt libraries