3–4 小时做出可玩小龙游戏：Images 2.5 起手
从 GPT Image 2.5 生成龙角色，经 Tripo3D 转 Three.js，再由 Astra 补骨骼动画与关卡，全程约 4685 行代码。
Okay, I’m not even sure where to start. All of this is kind of incredible to me. From basically nothing to a playable mini-game in just 3 to 4 hours. I started by asking ChatGPT Image 2.5 to generate a game-style dragon character. I then took that image into Tripo3D, used Smart Mesh with an 8K texture, and exported the unrigged and unanimated GLB model. If you want to try Tripo3D yourself, you can use my affiliate link here: http://studio.tripo3d.ai?via=sbr999 If you end up buying anything through it, you’ll also be supporting me, so I’d really appreciate it :) From there, I used my own converter to turn the model into Three.js. And yes, what you’re seeing here is actually running entirely in Three.js HTML, not just displaying a GLB file. Then I gave it to Astra and asked it to create all the rigging and animations, including idle, flying, walking, running, and attacking. The first result was already pretty decent, but I pointed out a few issues and had it improve them a couple of times. Once that was working well, I created an enemy using basically the same pipeline: ChatGPT Images -> Tripo3D -> Three.js. I also created the trees using a similar process. Then I asked Astra to make the terrain more realistic and better fit the overall theme of the game. After that, I had it add grass and some additional environmental details. As a final step, I asked it to redesign the HUD and create a proper game-style main menu. And... wow. Basically the only things here that weren’t AI-generated are the sounds. I think? :D The whole thing took around 3 to 4 hours, ended up being just 4,685 lines of code, and the entire HTML file is only around 15 MB. Another interesting part is the usage. This entire test only used around 15-20% of my weekly 5x Pro quota. I’m not sure whether OpenAI adjusted the limits recently, or whether converting the original GLB into my Three.js format made that much of a difference. In my earlier testing, the Three.js version was significantly smaller than the original GLB, and Astra also completed the same kind of task noticeably faster with it. So that optimization might actually be saving a pretty significant amount of compute and usage here. From zero to a playable 3D mini-game in a few hours, 4,685 lines of code, around 15 MB total, and only 15-20% of my weekly quota used. That still feels kind of insane to me. I’ll put the link to the game in the comments if you want to try it yourself. #astra #gpt6 #openai