Let's share our tools and help each other make the best games possible!

Here's my setup for the jam:

Code: Claude in VS Code

Art: GPT Image 2

Music: Suno

Engine: LittleJS. It's been working amazing with Claude. I'm using the version in this repo that has extra AI tools: LittleJS-AI

What's everyone else planning to use? Drop your stack below. 👇

I built an open source procedural sound generation library called chirrp that I've been using for my own games https://crates.io/crates/chirrp

It has an mcp server and a couple of skills that I've tested with my own games with great results, so I'll probably be using it for this jam as well.

Code: Antigravity

Music: Lyria

Engine: Bevy

Art: Likely Kenney assets, depends on the theme, might use trellis to create additional assets if required.

CLI: Codex CLI linux latest Models: GPT 6 Astra med-max & GPT 5.6 Luna xhigh-max Music: Lyria latest Art: GPT Image 2 Engine: Godot

Code / Agents: OpenCode with GPT-6 and OpenCode Go plan
Planning / Review: ChatGPT
Art / References: GPT Image
Engine: Godot 4.7.x
3D: Blender + Blender MCP
SFX: External CC0 sources and Soundly
Music: External CC0 sources

I tried something similar for the last game jam with Godot, Blender and VS Code but couldn't get the workflow reliable enough, so I went for Three.js instead. I'm specifically hoping to experiment with a more agent-driven Godot + Blender 3D workflow this time. The SFX and music I currently would like to go for finding external third-party sources like OpenGameArt, FreeSound, Soundly, etc. Preferably CC0 license audio assets, would like to hear what you guys suggest on audio parts here if you want to.

LMMS for music!
and chatPT for the rest

OpenAI's Codex/Dot/ChatGPT Image generation so I can work on it remotely even when I'm AFK. Suno for music, and free assets from itch for sfx where suno fails to capture what I'm trying to do. Invideo is nice for cinematics if you need those - but you will need a premium sub there or you'll run our of credits in like 47.9 seconds lol. 

I have a game I built in Unity using only DeepSeek. More recently, I’ve been working with agents using Mimo v2.6 via OpenCode. I use the free versions of all of these tools.

Claude for code, narration drafts and the itch copy; NanoBanana for the Polaroid character images; GitHub Copilot in VS Code for inline suggestions; Gemini for prompt ideas. Sounds are from freesound.org, edited in Audacity, and the hazard cues are synthesised with the Web Audio API. Game: Final Girl: The Fraternity House, a browser horror strategy game - https://storytellermeeka.itch.io/final-girl-the-fraternity-house

Models

• Claude Opus, in Claude Code: wrote all the code, levels and art. One lead session ran parallel helper agents.
• Meta MusicGen (musicgen-medium): the music.
• AudioLDM2: the sound effects. Both audio models ran through Hugging Face.

Tools

• The game: plain JavaScript, HTML5 Canvas 2D for graphics and WebAudio for sound. No engine and no build step.
• Testing: Node.js runs the test bot, and Playwright drives headless Chrome to replay the levels.
• Hosting: GitHub for the code, with Render auto-deploying every push. itch.io gets a zip uploaded by hand.

Techniques

• Levels: written as ASCII maps in a generator script.
• Graphics: almost all drawn in code, not image files.
• Determinism: the same inputs always play out the same way.
• Test bot: uses Wick's real physics to prove every level can be finished. It also sets the medal targets and the wax you respawn with at each torch.
• Replays: recorded runs are replayed in Chrome, and visual or sound changes must leave them exactly the same.
• Parallel agents: helper agents worked on different areas at once in separate copies of the repo, then their work was merged and checked.

Hello!! Here's the list of AI tools I used to make the game, with links in case they’re useful to you:

• Code: Claude Code, as an orchestrator coordinating a team of specialised agents (gameplay, UI, levels, art, sound and git).
• Art: PixelLab, connected to Claude Code through its MCP server.
• Music: ACE-Step 1.5, run locally.
• Sound: MOSS-SoundEffect v2.0, run locally with my own Mac adaptation.

I'm using mostly Codex / ChatGPT through the desktop app. Suno for music. Elevenlabs for sfx and voice (if I end up using any voice). Tripo mostly for models, have Meshy too but it's not as good. Astra and Sol 6.1 do fine and clean things up in Blender when needed.

Hi all! I use Claude Code with 2 MCP servers:

- Asesprite MCP for processing raw images

- locally hosted Stable Diffusion inference tool with an MCP server to generate the raw images for Asesprite locally

Claude Code does all of the coding, prompting of SD for images, and editing of the images in Asesprite, and I guide it via an agentic workflow I've set up using beads.

My stack is about as small as it gets: one AI, no engine, no asset files.

Code: Claude (Anthropic), just the regular Claude app chat. No IDE, no engine: plain JavaScript and HTML5 canvas in a single file.
Art: drawn in code by Claude. Every appliance, worm, gnome and boss is canvas drawing, with no image generator and no sprites.
Sound and music: synthesized live in the browser by Claude’s code (Web Audio), with a different lead instrument for each world. No Suno, no samples.
Testing: me on an iPhone and a laptop, plus little bots Claude wrote to play the levels and check the difficulty.
My job: director and playtester, over 140+ versions.

The result is AI WORMS, a one-button whack-a-mole through 11 worlds: https://seymour8glass.itch.io/ai-worms. I’d love to know what you think!

Adding a 3D-prop workflow from Quady3D AI, the company behind Quady3D:

Reference image → Tripo H3.1 GLB → inspect geometry and materials before engine import. One saved vehicle test contains 281,282 triangles in a 9.47 MB GLB. The silhouette is recognizable, but the panels look softer; engine performance is still unverified. We'd budget time for cleanup and simplification before repeating it across a scene.

Quady3D brings model selection, previews and file workflows into the browser. Generation uses credits, and full AI model downloads require a paid plan or credit pack:

https://quady3d.com/image-to-3d?utm_source=itchio&utm_medium=community&utm_campaign=ai_jam_prop_workflow

Disclosure: Quady3D is our product; the model is AI-generated and this note is AI-assisted. Are you mainly making new props or reworking materials on existing ones?

Our stack for Cream's Cozy Cottage:

Code and iteration: Codex.

Art: OpenAI image generation.

Engine: Godot, exported for desktop browsers with WebGL 2.

English localization: AI-assisted.

The demo combines material merging, crafting and cat care. Our onboarding work focuses on making the next action concrete: merge matching materials, craft a useful item, then use it to care for Cream. We also wrote a short first-treat guide for players.

We continued an earlier prototype; the jam entry describes that history. It is still an in-development demo, and mobile compatibility is not verified.

Entry: /jam/ai-jam-5/rate/5217300

This process note is AI-assisted too. We'd appreciate feedback on whether the first-care instructions make the next action clear.

Here's my stack for Kiosk Days: St. Martin's Evening:

• Code: Claude Code (Claude Opus). It writes all the GDScript, runs headless checks and even made the screenshots for the itch page.
• Engine: Godot 4.7, exported for web and Windows.
• Art: no image generator. Claude draws every sprite with Python and Pillow, pixel by pixel: the kiosk, the shelves, the lantern kids, St. Martin on his white horse, the fire. Everything comes from one shared library, so the style stays consistent.
• Sound and music: mostly generated in Python with numpy (the marching band, hooves, crowd murmur, crackling fire), plus a few free CC recordings.
• Balancing: a small simulator that plays many days headless, so prices are measured instead of guessed.
• My job: ideas, decisions and a lot of playtesting on PC and on my phone.

Funny part: Claude can't hear, so every sound goes through me before it goes into the game.

👉 https://kigaro-games.itch.io/kiosk-days-st-martins-evening

Hey everyone! 👋

I'm SihleM, an independent developer working under Poison Games. I'm participating in AI Browser Game Jam 5 with my game, Beneath the Borrowed Light, a browser-based survival mystery adventure inspired by the TORCH theme. 🔥

For this project, I've been using three main AI tools:

🤖 OpenAI ChatGPT

• Game design discussions and planning.
• Narrative development and writing assistance.
• UI/UX planning and refinement.
• Debugging discussions, testing strategies and quality assurance.

💻 OpenAI Codex

• AI-assisted programming using HTML5, CSS and JavaScript.
• Gameplay implementation and integration.
• Code refinement, debugging and optimization.
• Supporting functional and regression testing.

🎨 OpenAI Image Generation

• Original game artwork and visual assets.
• Cover art, branding and promotional materials.
• Visual experimentation and refinement.

One of the biggest lessons from this jam has been that generating code or artwork is only part of game development.

Getting everything to work together consistently, identifying issues and ensuring the game is actually enjoyable requires considerable iteration.

I'm curious to see what tools everyone else is experimenting with and what challenges you've encountered!

Good luck with your entries! 🔥

entry:https://sihlem.itch.io/beneath-the-borrowed-light

— SihleM | Poison Games

using Claude opus 5.5 in blender and godot 

Made the music myself and recorded skateboarding sounds in my garage 

Had ChatGPT mock up a reference photos for characters for opus to create in blender as well as deck designs 

I use GPT Codex and Claude to control https://www.makko.ai/ to create all the graphics and 2D assets and animations I need for the game. Codex/Claude act as the orchestrators and coders and Makko's Collections feature ensures that my characters, environments, props, and animations all remain consistent and look like they belong in the game game.