Nine years in the games industry building character pipelines for studios. Now I'm building one as a product: creategamecharacters.ai — game-ready 3D characters created in the browser. Face from a photo, AI-generated clothing fitted and rigged, motion capture from a webcam, one export button.
three.js · JavaScript · Node.js · Cloudflare · Supabase · Claude Code
The character pipelines I spent a career building inside studios, rebuilt as a product anyone can open in a browser tab. Pick a base, match its face to a photo or a text description, sculpt it, give it hair, dress it in AI-generated clothing that gets built into 3D and rigged to move with the body — then export a rigged, game-ready GLB. It's live, it's early, and it ships every day.
The parts I care most about:
- Face from a photo. Upload a photo (or just describe someone) and the editor matches the base head to it — landmarks, sculpt, skin — entirely in the browser.
- The Clothing Factory. Generative AI creates the garment from text or an image; the factory builds it into 3D, fits it to the body, and rigs it so it moves with the character.
- The Animation Tool. Motion-capture a performance straight from your webcam and refine it with human/AI-agent co-op animation tools.
- Agent-native by design. The site ships a documented machine API —
/agent.mdplus a verb manifest at/.well-known/ai-tools.json. An AI agent can list saved characters, create one from a photo, dress it and export the GLB end-to-end by URL, without clicking a single button. - Characters are numbers, not files. In the SDK a character is a few hundred bytes of slider weights on one shared base mesh — a crowd costs what one character costs. An in-game character creator calls
Creator.open()instead of rebuilding the logic. - Local-first. Your characters and outfits live on your device, in your folder. The only cloud surface is the community hub, where publishing is a deliberate act.
And the meta-layer, which is half the fun: the codebase itself is built AI-first. Multiple Claude Code sessions work the same checkout in parallel, routed by a per-folder context tree instead of search, and the public agent docs are generated from the internal ones. Building the product and evolving how agents build products are the same project.
Nine years at Attensi, where I was Lead Technical Artist and character pipelines were my whole job:
- The in-house Character Creator — I designed and built the Unity character creator that shipped customized characters to some of the largest companies in the world, before generative AI, running well on phones, tablets and low-spec machines. Customers kept calling it fun, which for a 3D production tool is the review that matters.
- CC4 → GLB automation — automated the full path from Reallusion Character Creator 4 to optimized, game-ready GLBs, making high-volume character delivery viable for a small team.
- RealTalk — built the initial demo of LLM-driven freeform dialog in games, proved it well enough that the company committed to it, then handed it off cleanly to the team that grew it into the industry-leading product it is today.
That decade is why Create Game Characters exists: I know exactly what a character pipeline needs when it has to deliver at scale to real devices. What changed is the cost floor — generative AI has made entire classes of pipelines viable that simply weren't worth building before, and I'm building the one I always wanted.
I am not a formally educated software engineer — I'm a technical artist who learned to code because the problems in front of me needed solving. Nine years of shipping games taught me how to set up pipelines that deliver effectively and consistently in large-scale production.
What that means in practice:
- I specialize in using tools, scripting and — these days — teams of Claude Code agents to build production pipelines, rather than hand-writing every line. My job is the architecture, the context the agents navigate by, and the taste to reject what isn't good enough.
- I've been using LLMs, diffusion models, and 3D generators since the early days, in pipelines that make real money, not demos.
- I have great intuition for how to deliver successful large-scale 3D production pipelines.
Production
Platform
Generative AI Stack
Footnote: Klonode — AI-driven experiment in context routing
TypeScript · Svelte · Node.js · Claude Code · ICM
Small experiment, not a main project. Klonode auto-generates per-folder CONTEXT.md routing for a git repo so AI coding assistants load only the context they actually need — the same idea that routes the Create Game Characters codebase. The codebase is itself mostly AI-written; I'm running it as an open playground for agentic workflows. AI contributors especially welcome — pick up an issue and ship.
