Adobe patent shows AI text-to-3D world generation for Firefly

The US Patent and Trademark Office has published a batch of Adobe filings, one describing a method to build navigable 3D worlds from text descriptions. An AI model first creates a spherical 2D panorama, which a second process converts into 3D geometry viewable from multiple angles.

• The filing describes generating a 3D scene from a prompt like "a foggy Japanese garden at dusk"
• Pipeline: spherical 2D panorama, then 3D geometry built by a second AI process
• Other filings cover new camera angles from one photo and 3D garment previews
• A patent application does not guarantee a shipping product