程序化卡通场景生成对比

Guess the cartoon this scene is from. Opus 5.5 is great, but do not write GPT-6 off. This test asked for a fully procedural build of a rather famous scene, no imports, no generators. And here, to my eye, GPT-6 did it noticeably better. Both models ran on Max effort, both took about an 1.5h. Opus 5.5: 141 objects, 51-step build timelapse. GPT-6: 616 objects, 727-step build timelapse.