Wink Pings

Gemini 4 Carbon Made a Kintsugi Vase, but the Comments Are Waiting for the Same Prompt

Pankaj Kumar used a single prompt to get Gemini 4 Carbon to generate a 3D kintsugi vase that shatters and reassembles, and said its design taste is comparable to Opus 5.5. The comments raised more practical questions: a side-by-side comparison using the same prompt, and whether it can hold on to the style after a second round of edits.

Pankaj Kumar posted a Gemini 4 Carbon output on X. The prompt was very specific: make a 3D kintsugi vase, shatter it, and reassemble it with gold cracks.

In the video, fine cracks first appear on a white ceramic vase, which then breaks apart and reassembles. The interface has Pause, Reset, Orbit/Inspect, and Cinematic Tour buttons, showing that it is not just rendering an animation but also provides an interactive way to inspect it.

Kintsugi is the craft of repairing cracks with gold, and its emphasis is on preserving the traces of damage. Turning this concept into a 3D scene means the model has to handle not only the vase model and crack material but also understand the narrative of shattering, pausing, and reassembling. For ordinary users, it looks like a beautiful demo; for developers, it contains a combined challenge of modeling, animation states, materials, and UI controls.

Pankaj Kumar's assessment was direct: Carbon is comparable to Opus 5.5 here, unlike Kimi K3.1 Pro. He also said that so far he really likes Carbon's design taste.

The comments did not stop at 'looks good.' Atakan said he uses Opus 5.5 every day for browser animations, so he would really like to see the same prompt run side by side on both. His reasoning holds up too: from a single video, it is hard to judge just how comparable 'comparable' really is. Pankaj Kumar's reply was to wait for release.

Another question is closer to everyday use. AndroId useR asked whether this quality could be maintained if Carbon were asked to make a second round of edits. Pankaj Kumar replied that he had only run it once, and that is how it was in the arena.

That reply makes the limits of the demonstration clear. A one-shot result can show the model's aesthetic taste and engineering completeness under a specific prompt, but it cannot show whether it can withstand repeated revisions. In real projects, second and third rounds of edits are the norm: changing materials, pacing, camera shots, and interactions. Whether the model can still hold on to the original style is harder than making the first version impressive.

Design taste cannot be benchmarked, and the comments still prefer to ask for a side-by-side comparison first. After Carbon is officially released, the same-prompt comparison and second-round edit tests will be more informative than a single demo.

Original tweet: https://x.com/pankajkumar_dev/status/2114021863605322191

发布时间: 2026-10-25 00:31