Four Game Prototypes Built with GPT-6 Astra and Other Tools

Peter Yang shows a space shooter, train shooter, strategy level, and deck builder—and the iterations behind each.

PYPeter Yangホスト · Peter Yang
チャンネル
Peter Yang
公開日
2026年9月5日
動画
23:49
読了時間
3 分
スライド
5
原文
英語
YouTube で開く
スクロールすると動画は右下に小さくなって再生を続けます。字幕は閲覧言語に合わせて切り替わります。
閲覧言語2 言語で読めます
この記事はまだ日本語に翻訳されていないため、英語の原文を表示しています。
要約

30 秒でわかる

Peter Yang demonstrates four game experiments using GPT-6 Astra alongside tools such as Blender, Godot and image generation. He revises mechanics and difficulty after play testing, and his deck builder begins as a Claude prototype before moving to ChatGPT. His model ranking is a personal assessment; the comparisons involve different tools and levels of effort.

  1. 1Use tools suited to the projectYang uses Blender and Godot for 3D prototypes and a web-based approach with 2D artwork for the deck builder.
  2. 2The visual comparison changes multiple variablesEarlier space-shooter examples used Three.js, while the newer demonstration also used Blender and Godot.
  3. 3Prototype, play and give specific feedbackThe first shooter was not fun; Yang asked for stronger enemies, power-ups and a better boss.
  4. 4Balance is part of game designYang adjusted difficulty after testing the strategy game and finding later waves too hard.
  5. 5The deck builder began in another modelClaude produced an initial prototype before Yang moved the code to ChatGPT and refined the art and gameplay.

ポイント

  1. Begin with the player experience and use concept art to communicate the desired look.
  2. Build milestones and play-test: Yang’s first shooter was not fun, and the strategy level needed rebalancing.
  3. The model comparison is not controlled because earlier examples and the newer version used different tools.
  4. No Moat started in Claude and evolved in ChatGPT through 2D art, balance work and mobile testing.
  5. The examples range from tech demos to a more developed prototype and all involve human feedback.
全文

チャプターごとに整えた全文を読む

文字起こしのみ
01

Four game experiments

Yang introduces his setup and personal assessment.

Peter Yang demonstrates four game prototypes made with GPT-6 Astra and other tools: a space shooter, a moving-train shooter, a real-time strategy level and a roguelike deck builder. He calls Astra the best model he has used for games, a personal judgment rather than a controlled benchmark.

His setup includes ChatGPT, the open-source Blender 3D tool and Godot game engine, with MCP connections. He says he asked the assistant to install the tools and used a medium reasoning setting. Tool availability and plan limits in the video reflect that moment.

02

A space shooter and a train shooter

Concept art, milestones and iterative feedback shape two prototypes.

The first prototype, Starfall, is inspired by Star Fox: the player flies through a collapsing structure, fights enemies and reaches a boss. Yang compares it to earlier model examples but notes that the older versions used Three.js while the Astra version also used Blender and Godot, so the comparison changes more than the model.

A space-shooter prototype
スライド 1 / 5A space-shooter prototypeStarfall combines a collapsing structure, player ship and combat.

He starts with ideas and reference art, then asks for teammates, power-ups, harder enemies and a more interesting boss. The first version was not much fun. He says the agent can play-test with browser and computer use, and he iterates based on his own experience.

Dustline is a first-person shooter on a moving train, with multiple weapons, enemies and a final boss. Yang calls it a tech demo. He asks for ideas, creates concept art, divides the work into milestones and provides feedback on visuals, enemy behavior and weapons, including by voice from his phone.

A shooter on a moving train
スライド 2 / 5A shooter on a moving trainDustline is a first-person tech demo with multiple weapons.
03

A real-time strategy level

Yang tests unit mechanics, lighting and difficulty.

Ashfall is a StarCraft-inspired base-defense level with controllable units, turrets and waves of enemies. Yang asks for specific mechanics, better lighting and difficulty adjustments. His own play test reveals that later waves became too hard, so he asks for a rebalance.

A strategy defense level
スライド 3 / 5A strategy defense levelAshfall gives the player units, defenses and enemy waves.
04

A roguelike deck builder

A prototype evolves through artwork, balance and mobile testing.

The fourth project, No Moat, is a roguelike deck builder about an AI founder. Players choose characters and companions, play cards and face encounters. Yang considers it closer to an actual game than the other prototypes because he spent more time on mechanics and balance.

The No Moat deck builder
スライド 4 / 5The No Moat deck builderThe player uses cards and characters to navigate encounters.

He began the deck builder with Claude before Astra was available, using a specification and an early emoji-based prototype. After moving the code to ChatGPT, he focuses on one convincing fight and generates two-dimensional character and environment art. An attempt at 3D characters is less satisfactory for his chosen style.

Generating game artwork
スライド 5 / 5Generating game artworkYang reviews generated 2D characters and assets for the deck builder.

Yang adds enemies and characters, tests the game, asks for balance and improves the mobile layout. He publishes the web game and says it took about two hours of back-and-forth. He observes that some generated character art relies on stereotypes.

05

Lessons and launch commentary

Yang summarizes his process and comments on launch availability.

His general process is to brainstorm an experience, provide concept art and iterate through play testing and feedback. The prototypes involve different tools, time investments and human decisions; his examples do not establish that any model reliably makes a finished game in one prompt.

Yang also discusses criticism of the Astra launch, arguing that announcements should coincide with broad availability. His suggestion that third parties may have affected release timing is speculation. He encourages viewers to experiment with their own ideas.

同じチャンネルの記事

46:40EN
Peter Yang· 2026年11月8日ポッドキャスト · 2 人Ashley Alexander: How AI Helps Me Care for My Parents With Cancerゲスト Ashley Alexander · ホスト Peter YangA caregiver uses AI to prepare for appointments, organize records and ask better questions—while keeping doctors at the center of treatment.AI in healthcareHuman-AI CollaborationProductivity
17:50EN
Peter Yang· 2026年11月4日ポッドキャストPeter Yang’s Five Steps for Using AI to Help Care for His MotherYang uses a shared care brief and organized records to navigate appointments, while making time for family.AI in healthcareHuman-AI CollaborationProductivity
15:03EN
Peter Yang· 2026年11月3日ポッドキャストHow Peter Yang Used an AI Phone Agent to Negotiate His Internet BillAn AI call negotiated a lower price and faster service, but Yang still had to approve the change himself.AI agentsProductivityHuman-AI Collaboration
33:09EN
Peter Yang· 2026年11月1日ポッドキャスト · 2 人Jaclyn Konzelmann: When AI Can Build More, Choose What Deserves to Existゲスト Jaclyn Konzelmann · ホスト Peter YangGoogle Labs pairs short strategy documents with prototypes, user co-creation and lessons from experiments that end.AI Product ManagementProduct StrategyHuman-AI Collaboration+1
スライド 4 枚13:05EN
Peter Yang· 2026年10月28日ポッドキャストPeter Yang Tests Genie 4: Convincing Worlds, Unfinished GameplayFive live demonstrations show scenes that respond to prompts and sound, alongside inconsistent actions and limited game mechanics.Generative media画像と動画
スライド1:06EN
Peter Yang· 2026年10月26日ポッドキャストPeter Yang on Setting Up Payroll for a One-Person BusinessIn a paid partnership, Yang describes the paperwork he encountered and a payroll product integrated with a business account.EntrepreneurshipProductivity