WHAT NEW IN X11.07 SAT

X 上的 AI 合集档案 · 23 个合集 · 70,791 条

gemini-4IMAGE

3D 任务实测:Argon 仍落后 Opus 5.5

对比 Google 官方 Argon 3D 演示与 Opus 5.5 一天内做出的版本,认为 Argon 在真实感和物体摆放上仍明显落后。

Leon Lin
@LexnLin21 天前收录原推发布于 2026.10.16
3D 任务实测:Argon 仍落后 Opus 5.5(第 1 张,共 4 张)
3D 任务实测:Argon 仍落后 Opus 5.5(第 2 张,共 4 张)
3D 任务实测:Argon 仍落后 Opus 5.5(第 3 张,共 4 张)
3D 任务实测:Argon 仍落后 Opus 5.5(第 4 张,共 4 张)
Leon Lin @LexnLin原推发布于 2026.10.16
在 X 上打开原推

Opus 5.5 is still WAY better than Argon at this 3d task. As you can see in the demo that Google shared, there are A LOT of things that either look very unrealistic, sloppy, or misplaced. They used some complex stuff, reference images, loops, multiple agents with visual critique, skills, etc. to create it, probably over many many hours, maybe even days. Whereas Opus made the one in the quoted video in less than a day, with a simple prompt + image references and it just nailed the realism better imo. You can clearly see that some things are placed incorrectly in the images here (from Google's demo), whereas in the Opus one, it knew exactly where things belong. And the trees looked more realistic, without leaves almost floating around, etc. So overall, Argon is a big improvement, but still a lot behind Opus 5.5 in 3D.

这条是引用着另一条写的:「Opus 5.5 用 Three.js 生成黄金时刻日本寺庙」

x.com/lexnlin/status/2110851952934293849采集于 2026.10.16 10:11