Remove Distractions from Portraits Using AI in Photoshop

Remove Distractions from Portraits Using AI in Photoshop

Almost every portrait has something in it you wish was not there. A sign in the background, a bright patch pulling the eye away from the subject, a stray hair across a cheek, a piece of litter on the ground. These are small problems that used to cost real time, because cloning over them means matching tone, texture and grain by hand.

Generative Fill handles most of them in seconds, and the technique has one detail that makes the difference between a clean result and a soft, obvious patch: what you type in the prompt box. The answer is nothing at all.

Why This Beats Cloning

The Clone Stamp and the Healing Brush copy pixels from elsewhere in your own image. That works beautifully when there is a clean, similar area nearby to borrow from, and it falls apart when there is not — a distraction against a complicated background, or one overlapping several different textures, leaves you nothing good to sample.

Generative Fill generates new content that fits the surroundings instead of copying existing content. It can reconstruct a background that never existed in the frame, which is precisely the case that used to take the longest.

Remove Distractions From Portraits: the Workflow

1. Press F for full screen

   Not essential, but it gives you the whole image without panel clutter, which makes distractions much easier to spot.

2. Press L for the Lasso Tool

   The standard Lasso. Nothing more precise is needed for most of this.

3. Draw around the distraction

   Include a margin of clean surroundings on all sides.

4. Click Generative Fill and leave the prompt empty

   Do not type anything.

5. Click Generate

   The selection is processed in the cloud and three variations come back.

6. Choose the best variation in the Properties panel

   Then move to the next distraction. One at a time, each on its own layer.

The Empty Prompt

This is the thing most people get wrong. The prompt box invites you to describe what you want — “clean wall”, “grass”, “sky” — and doing so tells the AI to generate that thing. Generated grass is rarely as convincing as the grass already in your photograph.

Leave the box empty and Photoshop interprets it as a removal instruction. It extends what genuinely surrounds the selection rather than inventing something new, which is almost always the better result when your goal is for something to simply not be there.

Save prompts for when you actually want to add something. For taking things away, type nothing.

Making the Selection

Rough is fine, away from the face

For a distraction in an open background, precision buys you nothing. A quick loose lasso with a generous margin gives the AI more surrounding context to match against and produces a better blend than a tight tracing would.

Tight is essential, near the face

This is the important exception. When a distraction overlaps or sits close to a facial feature, keep the selection as close to the object as you possibly can.

Anything inside your selection is territory the AI is allowed to rebuild. Include part of an eyebrow, a lip, or the edge of an eye, and it will reconstruct that too — plausibly, but not as the person’s actual feature. The result is subtly not them, which is far worse than the distraction you were removing.

Zoom in and work carefully anywhere near a face. Elsewhere, be quick.

Hold Shift to add

If a single distraction is split into pieces — an object partly hidden behind the subject, or a mark broken across two areas — hold Shift and draw around each piece. They become one selection and are generated together, which keeps the result consistent between them.

Holding Alt or Option subtracts, which is how you carve a face or a hand back out of a selection that overshot.

Texture and Grain

One of the better things about this approach is that generated areas pick up the existing grain and texture of the photograph rather than coming back artificially clean. On most images the patch sits in without any further work.

On very high-resolution files it is worth checking at 100% anyway. Generated content is capped in resolution, so a large removal can come back slightly softer than its surroundings. If it does, adding a small amount of noise to that layer settles it — and selecting smaller areas in the first place mostly avoids the problem.

Working Small: Stray Hairs

Individual stray hairs are worth a mention because they are the most common portrait annoyance and they respond well to this.

Zoom in properly — 200% or more — and lasso a narrow band along the hair rather than a large blob around it. A thin, close selection gives the AI a small, well-defined job and the results are consistently good. A big loose selection around a thin hair asks it to rebuild a large area of skin or background for no reason.

When One Pass Is Not Enough

Generate again

The three variations are genuinely different. If none work, clicking Generate again with the same selection produces three more, and often one of those is right.

Change the selection

If several rounds all fail, the selection is the problem rather than the AI. Widening it slightly, or shifting it to include more clean surroundings, usually changes the outcome more than rerolling does.

Stack passes

For a stubborn removal, run a second pass over what remains. Because each generation lands on its own layer, you can build the fix in stages without losing any of the intermediate steps.

Knowing What to Remove

Things that pull the eye

Bright patches near the frame edge, high-contrast objects behind the subject’s head, litter, signage, distracting highlights. Removing these is straightforwardly improving the photograph, and nobody looking at the result would say anything was misrepresented.

Things that are part of the person

Skin texture, marks, scars, features. These are a different category, and the tool being fast does not make the decision for you. A portrait subject generally has a view about what they want smoothed away and what they do not, and it is worth asking rather than assuming.

The useful test: if removing it makes the photograph better, do it. If removing it makes the person different, ask first.

For portrait retouching taught in order, our PRO Portraits Learning Path is where to go next.

Quick Reference: Keys, Menus and Settings

• Spot the distractionsF for full screen
• SelectL for the Lasso Tool; loose in the background, tight near the face
• Combine piecesShift-drag to add; Alt/Option-drag to subtract
• RemoveGenerative Fill in the contextual task bar, empty prompt, Generate
• ChooseCompare the variations in the Properties panel; select leftovers and generate again
• Before and afterGroup the generative layers and toggle the group