Prompt for Editing a Photo with AI

How to give an AI clear editing instructions

AI photo editors invert the old workflow. Instead of learning layers, masks and selection tools, you upload an image and describe the change in words. The catch is that plain language is easy to be vague in, and vagueness is the enemy of a good edit. "Make this look nicer" gives the model no target, so it changes whatever it wants — often the very things you were happy with. The fix is to be specific and to separate each edit into its own instruction, the way the prompt above uses a numbered list. "Brighten the photo," "remove the trash can on the left," "change the wall from beige to sage green," and "blur the background slightly" are four clear, checkable requests. A numbered list also makes it obvious when the model missed one.

Just as important is telling the AI what to leave alone. Add "leave everything else untouched" and, for photos of people, "keep the subject sharp and undistorted, no reshaping." Without those guardrails, a request to brighten a photo can quietly smooth someone's skin or shift their features. State the boundaries and the model stays on task.

Lighting, objects and color — what edits actually work well

Some edits are AI's strong suit. Lighting and color adjustments — brightening a dim shot, warming or cooling the tone, boosting or muting saturation, fixing a color cast — are reliable and look natural. Removing objects or people works well when the background behind them is simple; the model fills the gap convincingly. Background changes, like blurring for depth or replacing a cluttered scene, are also a sweet spot. Swapping colors of specific things — a shirt, a wall, the sky — is straightforward when you name the object and the target color precisely.

The edits that need care are ones that touch a person's face or body, and ones that change perspective or add complex new objects. Here the model can distort proportions or produce something that doesn't sit right in the scene. The single most common tell of a bad AI edit is a lighting mismatch — a new object lit from the wrong direction, missing its shadow, or too clean compared to the grain of the rest of the photo. That is why the prompt explicitly asks the model to match light direction, shadows, reflections and grain. When those line up, even a substantial edit reads as real.

Iterating conversationally, and when AI beats manual tools

The biggest advantage of AI editing over Photoshop is the conversation. You rarely get the perfect result on the first pass, and you don't need to — you refine it. After the first edit, react to what you see: "a bit too warm, pull it back," "you also removed the sign, put that back," "make the sky a deeper blue," "the shadow under that chair looks wrong." Each follow-up nudges the image closer while keeping what already works, which is why the prompt asks the model to show the result and then wait for your adjustments. Treat it as a dialogue, not a single command.

AI editing beats manual tools when the task is tedious or would need real skill — removing a distracting bystander from a holiday photo, evening out messy lighting, cleaning up a busy background, or trying several color treatments quickly. Manual tools still win when you need pixel-exact control, when the edit must be provably faithful to the original (legal, journalistic or scientific use), or when you can't accept the AI subtly altering anything it wasn't asked to. For everyday photos, though, describing the change and refining it conversationally is faster than any menu — and this prompt gives that conversation a clear, controllable starting point.

Why this prompt works

Conversational AI editors like ChatGPT and Gemini are powerful precisely because they take plain-English instructions — but that freedom is also where edits go wrong. Vague requests ("make it better") let the model change things you liked; over-broad requests distort faces and perspective. This prompt fixes both by making you list specific, numbered edits, explicitly protecting the subject and the parts you want left alone, and turning the interaction into an iterative loop: edit, review, refine. Naming what should not change is as important as naming what should — and asking for matched light, shadow and grain is what keeps an edit from looking obviously pasted in.

How to customize it

• Replace the numbered brackets with your actual edits before sending — the more specific each one is, the better the result.
• Always add what should stay the same; naming what not to touch prevents unwanted side effects.
• Treat it as a conversation: do a first pass, then refine with short follow-ups rather than one giant prompt.

Prompt variations to try

Lighting and color only

Edit the attached photo — adjust only lighting and color, change nothing else. Brighten the shadows, fix the [warm/cool] color cast toward neutral, add gentle contrast, and make the colors look natural and true to life rather than over-saturated. Do not remove or add anything, do not touch the subject's face or shape, and keep the framing and resolution the same. Photorealistic result.

Remove a distraction

Edit the attached photo to remove the [person/object] in the [location]. Fill the gap so it blends seamlessly with the surrounding background — match texture, lighting and perspective so there's no visible patch or blur. Leave the rest of the photo, especially the main subject, completely unchanged. Keep it photorealistic at the original resolution.

Replace or clean up the background

Edit the attached photo. Keep the subject exactly as-is — same pose, face, edges and lighting. Replace or clean up the background: [blur it for depth / replace it with a plain neutral backdrop / remove the clutter behind the subject]. Match the subject's original lighting direction so it doesn't look cut out, keep the edges natural, and make the whole image photorealistic and consistent.

Common mistakes to avoid

• Being vague. "Make it look better" lets the model change whatever it wants. List specific, numbered edits — brighten, remove X, recolor Y — so it has a clear target.
• Not protecting the subject. Broad edits can smooth skin or reshape a face as a side effect. Add keep the subject sharp and undistorted, no reshaping.
• Forgetting to lock the rest. Without leave everything else untouched, fixing one thing often changes three others you liked.
• Ignoring light matching. New or removed objects give themselves away with wrong shadows or grain. Ask the model to match light direction, shadows and grain so edits blend.
• Trying to nail it in one prompt. AI editing shines through iteration. Do a first pass, review it, and refine conversationally instead of rewriting the whole request.

Frequently asked questions

What kinds of photo edits does AI handle best?

Will AI editing distort the person in my photo?

How do I fix an edit that's almost right?

When should I use Photoshop instead of AI?

Tip: replace the parts in [square brackets] with your own details before you send. The more specific you are — audience, tone, goal, constraints — the better the AI output.