Inpainting
Filling in missing or masked regions of an image based on surrounding context—remove objects, fill holes, extend scenes.
Think of it like
Using Photoshop content-aware fill but 100x more realistic.
Example
Mask a person out of a photo; inpainting reconstructs the background as if they were never there.
How it actually works
Diffusion model conditioned on masked image and mask. Learn to hallucinate content that matches surrounding pixels. Can also condition on text ("replace the person with a robot") for guided inpainting. Training uses images with random masks and content; inference provides mask and sometimes text.
For product teams
Enables photo editing at scale; remove logos, watermarks, or unwanted objects without hiring Photoshop experts.
For engineers
Masked image and mask concatenated with model input; diffusion conditioned on both; noise only diffused in masked regions.
Related
- Diffusion — Variant of.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome