Google Patents an AI That Finishes Cut-Off Objects and Moves Them Anywhere in a Photo
You know that great photo where the subject is half-cut off by the edge of the frame? Google is patenting an AI that not only figures out what the missing part looks like — it lets you move the completed object anywhere else in the shot.
What Google's object-completion photo tool actually does
Imagine you have a photo of your dog, but his tail is chopped off by the right edge of the frame. Or a friend is mostly hidden behind a lamppost. Normally, you're stuck with the photo as-is, or you'd need serious Photoshop skills to fix it. Google's patent describes a tool that handles this automatically.
The system lets you tap the cut-off or partially hidden object. It figures out what the missing portion probably looks like — filling in the gap using AI — and hands you back a complete version of the object. At the same time, it cleans up the original background behind where the object was sitting.
The really useful part: once the AI has produced a complete version, you can place it anywhere in the photo — not just where it was cropped. So your dog with the full tail could be repositioned to the center of the frame, blended naturally into the background.
How the diffusion model fills in and relocates objects
The patent describes a pipeline with several distinct steps that work together:
- Object selection: You pick the incomplete object — one that's either cut off by the photo's edge or hidden behind something else.
- Mask generation: The system creates an "object mask" — essentially a precise outline of where the incomplete object sits, pixel by pixel.
- Inpainting the background: The original object pixels are erased, and an AI fills in what the background behind the object probably looks like — a process called inpainting (think of it as an intelligent fill that invents plausible scenery).
- Object completion: A diffusion model (the same class of AI that powers image generators like Midjourney or Google's own Imagen) takes the incomplete object, the mask, and the cleaned background as inputs, and outputs a completed, whole version of the object.
- Blending: The completed object is composited back into the inpainted image at a new location the user chooses, using the mask to make the edges look natural.
The claim specifically notes that the final position can be different from where the object originally sat — meaning this isn't just a repair tool, it's a repositioning tool.
What this means for everyday photo editing on Google products
For everyday users, this collapses what used to be a multi-step, skill-heavy editing job into a single tap-and-place interaction. Fixing a badly framed photo — or rescuing an otherwise unusable shot — becomes something anyone could do.
Google Photos is the obvious home for this. The app already uses AI for Magic Eraser and other automated edits, and this patent fits squarely in that trajectory. It also has implications for Google's Pixel camera ecosystem, where computational photography features are a key selling point against Apple and Samsung.
This is a genuinely useful editing capability, not a vanity AI demo. The combination of object completion and repositioning in one workflow solves a real, common problem — badly framed photos — rather than inventing a use case nobody asked for. If it ships in Google Photos, a lot of people will use it weekly.
Get one Big Tech patent every Sunday
Plain English, intelligent commentary, no hype. Free.
Editorial commentary on a publicly published patent application. Not legal advice.