GD Studio and Art Factory: how a small team uses AI tooling to design and produce faster

Challenge

• 1

  A studio shipping a new puzzle title every few weeks needs a steady supply of two things: well-scoped game design documents that target the right KPIs, and consistent, Unity-ready art. Both are usually bottlenecks for small teams.

• 2

  Off-the-shelf AI tools helped individuals but not the team: prompts lived in personal chats, art styles drifted between artists, and nothing connected design intent to the numbers that decide a prototype's fate.

Objective

1. 1

   Give every designer the same expert assistant, tuned to the studio's four north-star KPIs: D1 and D7 retention, US CPI and playtime.

2. 2

   Turn a concept into a full GDD skeleton and a KPI estimate in minutes, in a workspace the whole team shares.

3. 3

   Produce game art from structured prompts with style presets and reference libraries, run technical QA automatically and export assets straight into Unity.

Approach

1. 1

   GD Studio (gd.cmzsoft.com) wraps a large language model in seven game-design 'skills' — core loop, retention, level design, monetization, UA/creative, meta & economy, playtest — each a system prompt written to pull a KPI in the right direction, with a KPI predictor and an idea generator on top.

2. 2

   Art Factory (art.cmzsoft.com) is a Next.js + worker platform: projects, style presets, a reference library, a prompt engine that composes structured prompts, edit/version history, automated technical QA (size, alpha, naming) and a one-click Unity export package.

3. 3

   Both tools are bilingual (Vietnamese/English) so the whole team works in their own language while game-design terminology stays consistent.

Execution

1. 1

   GD Studio runs as a React application with serverless functions calling the Claude API; conversations, GDDs and analyses are stored in a shared team workspace with author, search and Markdown export.

2. 2

   Art Factory runs a Postgres/Supabase backend, a job queue and a processing worker; artists generate, edit and approve assets, and the Unity importer package pulls approved exports into the project.

3. 3

   Both feed the same production loop: a concept goes GD Studio → prototype on the puzzle framework → art from Art Factory → test launch on Google Play → analytics readout.

Result

• 1

  Design and art stopped being the pacing items for new prototypes; the constraint moved to test-launch capacity, which is the right place for it.

• 2

  Every concept now arrives at the go / kill meeting with a KPI hypothesis written the same way, making prototypes comparable.

• 3

  Art direction stays consistent across sixteen titles produced by a small team, and assets arrive in Unity already named, sized and QA-checked.

• 4

  The tools are internal products with their own roadmaps — the studio treats production tooling as part of the product, not a side project.

What we learned

• 1

  An assistant is only useful to a team when its output lands in a shared place with a name on it.

• 2

  Encode the KPIs into the tool. A prompt that knows the D7 target writes different level-design advice than a generic one.

• 3

  Automated QA on art is worth more than better generation — consistency and Unity-readiness are what save production hours.