Microsoft Copilot tests turning 2D pictures into 3D models

Anyone who has ever touched 3D modeling knows the drill: spending hours tweaking vertices just to get a basic shape right. Microsoft is now trying to cut that process down to a single prompt inside Copilot.

According to findings from TestingCatalog, Microsoft is developing a new Copilot 3D experiment that lets users turn regular 2D images into 3D models.

From a single photo to a 3D asset

The feature allows users to upload an image and have Copilot generate a corresponding 3D object directly inside the chat interface.

The update also brings a new Recreate button, which lets users duplicate pre-generated 3D items. Hitting that button takes you straight to a My Creations section where you can browse, preview, interact with, and download your past builds.

Microsoft seems to be aiming this directly at casual creators, game developers, educators, and digital artists who just need fast, lightweight 3D content without opening specialized software.

The fidelity hurdle and the Labs curse

The catch? Early testing shows that the generated 3D models still suffer from poor fidelity. The meshes are rough, signaling that Microsoft still has plenty of engineering work left before these assets are practically usable.

There is also zero word on if or when this experiment will leave private testing. The feature will likely live inside the experimental Copilot Labs hub, a testing ground that does not have the cleanest track record.

Past experiments, like Copilot Characters, were quietly deprioritized, stripped down to basic voice features, and left without updates for months. Copilot 3D could easily suffer the same fate if generation quality stalls or user retention stays low.

Why it matters

We think dropping 3D generation into a mainstream assistant like Copilot is a very clever distribution play. If you are an indie game maker or a concept artist, being able to prompt a rough gray-box asset in seconds saves real production time, even if the model requires cleanup later. That said, given the current fidelity issues, it is not worth hunting for access right now.

For the broader AI space, this pushes multi-modal assistants beyond text, code, and images into spatial assets. If Microsoft pulls it off, competitors like Google and OpenAI will face pressure to treat 3D as a standard output format.

The real danger is Microsoft's habit of abandoning its experimental side quests. 3D assets carry strict technical requirements around topology, textures, and poly counts that simple generative models struggle to meet. If Microsoft treats this merely as a fun Labs gimmick rather than building robust export pipelines, Copilot 3D will likely stay a toy rather than a serious creative tool.