Poly generates 3D textures from text prompts

Anyone who has ever built a 3D scene knows the nightmare of texture hunting. You spend forty-five minutes scrolling through asset stores for mossy cobblestone, only to settle on something that looks like plastic.

Poly, a fresh startup out of Y Combinator, thinks generative AI can kill that chore entirely. The platform lets creators generate 3D textures using nothing more than simple text prompts.

Alongside the launch, Poly announced a $3.9 million seed round backed by Felicis, Bloomberg Beta, NextView Ventures, and Figma Ventures.

Typing out textures with full depth maps

Instead of offering generic flat images, Poly targets video game developers, 3D artists, and AR/VR creators.

You type in a prompt like "tree bark with moss" or upload an image reference, and the tool outputs customizable textures ready for physically-based rendering (PBR).

Crucially, it generates normal maps and invert maps alongside the color layer. That gives 3D surfaces authentic volume, depth, and realistic light responses without requiring manual sculpting.

Co-founder Abhay Agarwal, a former Microsoft researcher whose previous studio helped build choreography algorithms for BeatSaber, says Poly trains its models using proprietary techniques that pull texture data directly from standard images.

Despite having just three employees, the platform has already generated more than two million textures, with thousands of developers on its free tier and hundreds paying for pro accounts.

The stock asset library, fully automated

Poly is essentially positioning itself as an Adobe Stock or Shutterstock where every single asset is conjured by machine.

Noncommercial use is currently free and unlimited. The company charges a subscription fee for commercial and resale rights, with plans to expand into 2D sprites, illustrations, and sound effects down the road.

Of course, the timing comes with plenty of friction. While Poly barrels ahead, traditional platforms like Getty Images have banned AI art outright, and artists across communities like ArtStation have actively protested automated generators scraping their portfolios.

Agarwal insists Poly is meant to empower creators rather than replace them, and claims the startup has a multi-year runway to refine its tools.

Why it matters

Poly picked a remarkably smart wedge into the generative AI boom. While hundreds of clones battle over generic 2D image generators, 3D surface texturing is an unglamorous, highly technical bottleneck that creators actively hate spending hours on.

For indie game developers and small architecture studios, this is an immediate win. Being able to spit out workable PBR textures with pre-calculated normal maps in seconds dramatically shrinks pre-production timelines and removes the need to buy generic asset packs that every other indie game uses.

However, commercial adoption by larger gaming studios will face real hesitation. Between ongoing copyright lawsuits against generative models and backlash from human artists, many enterprise teams will tread lightly before baking AI-generated textures into major releases. Poly's survival depends on embedding so deeply into professional workflows that the productivity gains outweigh the legal ambiguity.