Pros almost never make a picture in one shot. They generate, then edit — paint over a spot they don't like and let the AI redraw only that patch so it blends in. Grab a brush, scribble over part of the picture, and watch a fresh patch grow to fill the hole.
Drag your finger/mouse across the picture to mask a patch (it glows magenta). Then hit Regenerate patch: only that spot gets erased to static and re-grown from the colors around it — the rest of the picture never moves. That's how AI editing really works.
Tiny mask = seamless, magic fix. Big mask over the main thing = the AI has to invent from the edges… and it guesses. Try both!
Go back up and paint a big mask right over the most important object (the sun, the face, the house). Hit Regenerate. See how the fill turns into a smeary blob that doesn't match? Inpainting only knows the surrounding pixels — it can't see what used to be there, so it invents a guess from context. Small blemishes vanish; big holes get made up.
Inpainting fills a hole inside. Outpainting does the opposite: it adds a fresh border outside and lets the AI keep the scene going. Press Extend a few times and watch the frame grow — the new edges are generated from whatever was touching them.
Notice the new border continues the colors at the edge — but it's inventing, so it gets vaguer the farther out it goes. Real tools like Photoshop's "Generative Expand" do exactly this, way sharper.
Each photo has an ugly blemish stamped on it (the glowing splotch). Paint a mask that covers it, then hit Fix it. If your mask covers the blemish, generative fill rebuilds clean pixels from around it and it's gone. Clean up 3 photos to win the badge!
Cover the whole splotch — miss a corner and the fill leaves a smudge, just like real life. 🩹
The mask-and-regenerate move you just did is shipping in the tools artists use every day:
Select a region, type what you want, and it inpaints it — add a hat, swap the sky, erase a tourist.
Phones now erase people and objects by masking them and inpainting the background. Same idea, one tap.
Turn a square photo into a wide banner by growing the edges. The AI dreams up what was "off-camera".
Hand-painting is fun, but pros let AI find the region too. This loads a real segmentation model (SegFormer) that runs on your device. It looks at the Inpaint Studio picture, labels every area (sky, tree, person…), and you tap one to auto-mask it — then Regenerate up top.
First load downloads from Hugging Face, then it's cached offline. On a 4GB laptop give it a minute. ☕ Totally optional — the brush above already teaches the whole idea.
You stopped generating and started editing. You masked a region and let AI redraw only that patch (inpainting), grew a picture past its edges (outpainting), and learned the catch: there's no hidden original — the fill is a smart guess from the neighbours.