Content-Aware Fill in Photoshop: When the AI Guess Is Good Enough (and When It Isn't)

You've got a great shot ruined by one thing: a trash can in the corner, a stranger walking through the background, a power line slicing across the sky. The old fix was an hour with the Clone Stamp. Since Photoshop CC 2019, the faster answer is Content-Aware Fill — a dedicated workspace that samples the pixels around your selection and synthesizes a plausible replacement. The thesis of this piece is simple: it's excellent, but only when you understand what it's actually doing and where it breaks down.

What Content-Aware Fill actually does

Content-Aware Fill is not magic erasure. It uses Adobe's Sensei AI to analyze textures, colors, and patterns in a sampling area — the green overlay you see in the workspace — and generates new pixels that statistically resemble that source material. This distinction matters: it remixes what already exists in your image. It cannot invent a brick wall it has never seen, and it will happily copy a distraction from one side of the frame into your fill if that distraction falls inside the sampling area.

That's why the dedicated workspace (introduced in CC 2019, replacing the older one-shot dialog) is the real feature. It gives you a live preview and, critically, control over what the algorithm is allowed to look at.

A worked example: removing a hiker from a landscape

Suppose you have a 24-megapixel landscape photo with a hiker standing on a ridgeline against sky and grass. The workflow:

1. Duplicate your background layer (Ctrl/Cmd+J) so you have a fallback.
2. Grab the Lasso tool (L) and draw a loose selection around the hiker — include a small margin of surrounding pixels, but not so much that you pull in unrelated elements like a nearby tree.
3. Go to Edit > Content-Aware Fill. The workspace opens with a green sampling overlay, a preview pane, and fill settings.
4. Use the Sampling Brush to subtract anything you don't want sampled — for example, if the hiker's shadow or a rock formation is inside the green zone, paint it out.
5. In the Fill Settings panel, nudge Color Adaptation up if there's a lighting gradient across the area (common with skies), and try Rotation Adaptation if the background has repeating angled structures.
6. Under Output Settings, choose New Layer (or enable the fill-layer option) so the result lands non-destructively, then click OK.

Zoom to 100% and inspect the fill edges. A good result on sky-and-grass should be nearly invisible; if you see repeating "cloned" tufts of grass, undo and restrict the sampling area further.

The settings that actually matter

Most defaults work, but two settings rescue difficult fills:

• Color Adaptation: blends the fill with gradual color and brightness changes. Essential for skies, walls with light falloff, and water.
• Scale and Mirror: useful when filling patterns (foliage, fabric) where mirroring the source texture avoids obvious repetition.

The sampling area, though, is the highest-leverage control. A tighter, cleaner sample beats any slider adjustment.

Where it falls apart

Be honest with yourself about these cases before you start:

• Low-contrast or complex edges: an object overlapping a face, text, or a detailed architectural edge will produce smeared, dreamlike artifacts. Manual retouching wins here.
• Structured geometry: the algorithm doesn't understand perspective lines. Removing something from a tiled floor or a fence often bends the lines subtly — viewers notice even when they can't say why.
• Big fills on big files: on a high-resolution image with a large selection, the preview can get sluggish on modest hardware. Working on a merged, downsized copy for the fill and then upscaling the result is a viable workaround, though it costs you some sharpness.

A practical closing habit

Always output to a new layer, and always do a 100%-zoom pass over the fill boundary before calling it done. If the result is 90% there, don't re-run the fill five times — a thirty-second touch-up with the Spot Healing Brush on the new layer is faster and more predictable. Treat Content-Aware Fill as a very fast first draft, not a finished retouch, and it will save you hours without embarrassing you.