These are 24 complete, paste-ready editing prompts for GPT Image 2 — OpenAI's most capable image model, built into ChatGPT and available via the API as gpt-image-2. They assume you've already attached a photo (and, where useful, a reference image). Its native Thinking Mode reasons about your existing image before it edits, so you don't generate from scratch — you describe the one thing to change.
Every prompt follows the core rule of conversational editing: state exactly what to change and what must stay identical. Naming what stays — the face, pose, lighting, and identity — is what keeps an edit believable, and telling the model to preserve the original aspect ratio and resolution stops it from cropping your image. For the full method, read the best GPT Image 2 prompts roundup and the guide to prompting GPT Image 2 for photorealism.
Background & scene
Four prompts for swapping, moving, removing, or blurring what's behind your subject. In every one, lock the subject so only the background changes.
1. Swap to a Grey Studio Backdrop
Using the attached photo, keep the subject's face, pose, hair, clothing, and lighting exactly the same. Replace only the background with a smooth, seamless mid-grey studio backdrop, softly lit with even three-point studio lighting and a gentle falloff toward the edges. Match the shadow the subject casts to the new backdrop so it looks physically real. Do not alter the subject in any way, and preserve the original aspect ratio and resolution.Why it works: Listing the face, pose, hair, clothing, and lighting as "keep exactly the same" gives GPT Image 2 an explicit lock, so it re-renders only the pixels behind the subject.
2. Move a Subject to a New Outdoor Scene
Using the attached photo, keep the person exactly as they are — same face, expression, pose, outfit, and body proportions. Replace only the background with a sunlit city park path at golden hour, soft warm backlighting with long shadows. Relight the edges of the subject just enough to match the new golden-hour light, but do not change their features or pose. Blend the ground contact and shadow so they sit naturally in the scene. Keep the original aspect ratio and resolution.Best for: Rescuing a good portrait shot in a dull location — the subject moves scenes without becoming a different person.
3. Remove a Busy Background
Using the attached photo, keep the subject exactly the same in every detail. Replace the cluttered background behind them with a clean, solid off-white backdrop, evenly lit with no visible seam or gradient. Preserve the subject's edges, individual hair strands, and the original lighting on their body. Keep a soft, realistic contact shadow beneath them. Do not change the subject, and preserve the original aspect ratio and resolution.Why it works: A solid off-white backdrop is the reliable way to isolate a subject; preserving hair strands and the contact shadow is what keeps the swap from looking pasted-on.
4. Blur the Background for Depth
Using the attached photo, keep the subject perfectly sharp and unchanged — same face, pose, and lighting. Add a shallow depth-of-field blur to the background only, as if shot on an 85mm f/1.4 lens, with soft creamy bokeh on any highlights. Keep the transition at the subject's edges natural, including fine hair detail. Do not blur or alter the subject, and preserve the original aspect ratio and resolution.Best for: Giving a flat phone snapshot a portrait-lens look without touching the person in front.
Retouch & cleanup
Four prompts for cleaning up skin, light, and small distractions — the kind of fixes where "keep it natural" is the whole job. Name pores and texture as things to keep so nothing turns plastic.
5. Retouch Skin Naturally
Using the attached portrait, retouch the face. Keep the person's identity, bone structure, expression, skin tone, and pose exactly the same. Gently reduce temporary blemishes and calm redness, but keep natural skin texture, pores, freckles, and fine lines — do not smooth or airbrush the skin. Leave the eyes, hair, and lighting untouched. The result should look like the same real person on a good day. Preserve the original aspect ratio and resolution.Why it works: Naming pores, freckles, and fine lines as things to keep stops the model from plastic-smoothing the face, which is the usual failure mode of AI retouching.
6. Fix Harsh Flash and Shadows
The attached photo has a harsh on-camera flash. Keep the subject's face, pose, expression, and background exactly the same. Relight it to look like soft, diffused window light instead: reduce the blown-out highlights on the forehead and nose, lift the harsh shadows under the chin, and restore natural skin tone. Do not change the subject's features or the composition, and preserve the original aspect ratio and resolution.Why it works: Describing the target lighting in real terms — "soft, diffused window light" — plus naming the specific hotspots to fix gives the model a concrete relighting job instead of a vague "improve" request.
7. Whiten Teeth and Brighten Eyes Subtly
Using the attached portrait, keep the face, expression, skin, and lighting exactly the same. Make only two subtle changes: slightly whiten the teeth so they look natural, not artificially white, and gently brighten the eyes and catchlights. Do not change eye color, tooth shape, or any facial feature. Keep everything else identical, and preserve the original aspect ratio and resolution.Best for: A quick, honest polish on a headshot — two small edits, nothing else moved.
8. Remove Braces, Stray Hairs, and Distractions
Using the attached portrait, make a few small cleanups only: remove the stray hairs across the forehead, the small piece of lint on the shoulder, and the shine on the tip of the nose. Keep the person's face, identity, expression, hairstyle, skin texture, clothing, and lighting exactly the same. Do not reshape any feature or alter anything not named. Preserve the original aspect ratio and resolution.Why it works: Enumerating each tiny distraction and closing with "do not alter anything not named" keeps GPT Image 2 focused on spot cleanups instead of a full retouch.
Recolor & restyle
Four prompts for changing color and mood — a film grade, dramatic light, a new season, and a clean recolor — while the scene and subject stay put.
9. Restyle a Photo as Golden-Hour Film
Restyle the attached photo with a warm 35mm film look: soft golden-hour color grade, gentle grain, slightly lifted blacks, and warm highlights with a subtle halation glow. Keep the exact same subject, pose, composition, and scene — change only the color, tone, and grain, not the content. Preserve every face and object as it is. Do not crop or reframe, and keep the original aspect ratio and resolution.Why it works: Separating "change only the color, tone, and grain" from "not the content" tells the model this is a grade, not a re-shoot, so the scene stays intact.
10. Relight From Flat to Dramatic
Relight the attached portrait. Keep the subject's face, pose, expression, and outfit exactly the same, and keep the background scene the same. Change only the lighting to dramatic chiaroscuro: a single soft key light from the upper left, deep controlled shadows on the opposite side, and a subtle rim light separating the subject from the background. Keep skin tones natural, and preserve the original aspect ratio and resolution.Best for: Turning a flatly-lit snapshot into a moody portrait without moving the subject or the set.
11. Change the Season or Weather
Using the attached outdoor photo, keep the people, their clothing, poses, and the main structures exactly the same. Change only the season: turn the green summer trees into autumn oranges and reds, add fallen leaves on the ground, and shift the light to a cooler, low autumn sun. Keep the composition and all subjects identical. Do not add or remove people, and preserve the original aspect ratio and resolution.Why it works: Constraining the change to season while explicitly protecting people, clothing, and structures stops the model from re-inventing the whole scene around the new mood.
12. Convert to Rich Black and White
Convert the attached photo to a rich, high-contrast black and white with deep blacks, clean bright highlights, and a full tonal range, like a classic silver-gelatin print. Keep the exact same subject, pose, composition, and detail — change only the color to monochrome and adjust contrast. Do not crop, retouch faces, or alter the scene, and preserve the original aspect ratio and resolution.Best for: A proper black-and-white conversion with real contrast control, not a flat desaturation.
Restore & colorize
Four prompts for bringing old and damaged photos back — repair, color, sharpness, and corrected white balance. Ask for 2K when the detail has to hold up in print.
13. Restore a Damaged Old Photo
Restore the attached scanned old photograph. Repair scratches, dust, creases, tears, and fading, and rebuild damaged areas plausibly. Keep the people's faces, expressions, clothing, hairstyles, and the original composition exactly the same — do not modernize, reshape faces, or add detail that changes who they are. Keep it black and white if the original is black and white, and preserve the period look. Output at 2K, matching the original aspect ratio.Best for: Family archives — the identity lock keeps a restored ancestor recognizably themselves instead of a lookalike.
14. Colorize a Black-and-White Photo
Colorize the attached black-and-white photograph with natural, period-accurate color. Keep every face, expression, pose, clothing shape, and the composition exactly the same — add only color, do not redraw or reshape anything. Use realistic skin tones, believable fabric colors for the era, and natural ambient light. Keep it subtle and photographic, not oversaturated, and preserve the original aspect ratio and resolution.Why it works: "Add only color, do not redraw or reshape anything" keeps colorizing an overlay rather than a regeneration, so identities and shapes hold.
15. Sharpen a Soft, Low-Resolution Photo
Enhance the attached slightly soft, low-resolution photo. Increase apparent sharpness and recover fine detail in the eyes, hair, and fabric, and render the result at high resolution. Keep the subject's identity, features, pose, colors, and composition exactly the same — do not invent new details on the face or change the expression. Avoid over-sharpening halos and plastic texture. Output at 2K, matching the original aspect ratio.Best for: Getting a print-ready, crisper version of a photo that came out soft — with a guard against invented facial detail.
16. Fix White Balance and Faded Color
Correct the color in the attached photo. Remove the strong yellow-orange color cast, set a neutral white balance so whites look white, and restore the faded, washed-out colors to natural saturation and contrast. Keep the subject, pose, composition, and all content exactly the same — this is a color correction only. Do not retouch faces or change the scene, and preserve the original aspect ratio and resolution.Why it works: Framing the request as "a color correction only" and naming the exact cast to remove keeps the model from creatively re-editing the scene.
Portraits & headshots
Four prompts for professional headshots and identity-locked people edits — the edits people pay for most. A reference image keeps one face consistent across variants.
17. Professional LinkedIn Headshot
Using the attached casual photo, turn it into a professional headshot. Keep the person's face, identity, expression, hair, and skin texture exactly the same — this must clearly be the same person. Replace the background with a soft, out-of-focus neutral office-grey backdrop, and light the subject with a clean three-point softbox setup for even, flattering light. Keep their real clothing if it looks business-appropriate; otherwise leave it unchanged. Frame head-and-shoulders. Aspect ratio 4:5, 2K.Why it works: The line "this must clearly be the same person" plus keeping skin texture forces GPT Image 2 to elevate the photo without swapping in an idealized stranger's face.
18. Consistent Headshot From a Reference
Use the attached reference image for this person's exact face and identity. Create a professional headshot that keeps that face, expression, and skin texture faithful to the reference — same person, no idealizing. Dress them in a plain dark blazer over a white shirt, on a soft mid-grey studio backdrop with even three-point lighting, head-and-shoulders framing. Keep the face identical to the reference in every version. Aspect ratio 4:5, 2K.Best for: Building a matching set of headshots — the reference image anchors one face so every outfit variant is recognizably the same person.
19. Remove a Photobomber
Using the attached photo, remove the person in the background on the right side. Reconstruct the background behind them to match the surrounding scene naturally — same wall, floor, lighting, and perspective. Keep the main subjects in the foreground, their poses, faces, and lighting exactly the same. Do not alter anyone you are keeping, and preserve the original aspect ratio and resolution.Why it works: Pointing to the exact person ("background on the right side") and telling the model what to rebuild the gap with keeps the fill consistent with the real scene.
20. Swap an Outfit Color
Using the attached photo, keep the person's face, pose, hair, body, and background exactly the same. Change only the color of their jacket from blue to deep burgundy, keeping the same fabric texture, folds, seams, and the way light falls across it. Do not change the shape or style of the jacket, and do not alter anything else in the image. Preserve the original aspect ratio and resolution.Best for: Testing a color variant on a garment while the fabric's real texture and drape survive the swap.
Product & e-commerce
Four prompts for clean, sellable product edits — including a true transparent-PNG cutout. Naming the label and logo as "keep unchanged" protects the branding the model is most likely to garble.
21. Clean Up a Product on White
Using the attached product photo, clean it up for an e-commerce listing. Keep the product's exact shape, color, label text, logo, and proportions unchanged. Place it on a pure, evenly-lit white background with a soft, realistic contact shadow beneath it. Remove dust, smudges, fingerprints, and any background clutter. Do not restyle or redesign the product. Aspect ratio 1:1, 2K.Why it works: Naming the label text and logo as things to keep unchanged protects the exact branding — the detail image models are most likely to garble. See the product photography prompts for a full set.
22. Cut Out a Product to a Transparent PNG
Using the attached product photo, isolate the product and place it on a transparent background — output a PNG with a true alpha channel, no background fill and no drop shadow. Keep the product's exact shape, color, label text, logo, materials, and proportions unchanged, and preserve clean, precise edges including any fine or reflective detail. Do not add a backdrop or surface. Aspect ratio 1:1.Best for: Catalog and web assets — GPT Image 2 writes a real alpha channel, so ask for "transparent background, no background fill" (and set background:"transparent" with output_format:"png" in the API).
23. Remove Reflections and Dust From a Product
Using the attached product photo, remove the distracting window reflections and glare on the glossy surface, along with dust specks and fingerprints. Keep the product's exact shape, color, label text, logo, and material finish unchanged, and keep the existing background and lighting the same. Rebuild the cleaned areas so the surface looks naturally lit and consistent. Preserve the original aspect ratio and resolution.Why it works: Reflections and dust are localized fixes, so naming them specifically — while locking the label, finish, and background — keeps the retouch from flattening the product's real material.
24. Place a Product Into a Lifestyle Scene
Use the attached image as the exact product reference. Keep the product's shape, color, label, and logo faithful to the reference — do not alter the packaging or text. Place it naturally on a warm wooden café table beside a coffee cup and a folded newspaper, in soft morning window light with a shallow depth of field. Match the product's scale, perspective, and reflections to the scene, with a realistic contact shadow. Aspect ratio 3:2, 2K.Best for: Turning a plain packshot into a lifestyle image while the real product stays exactly as it ships.
Once these feel natural, keep the roundup of the best GPT Image 2 prompts open for the underlying method, and pair this with the product photography collection for the shots you'll clean up here.
Frequently Asked Questions
What is conversational editing in GPT Image 2?
You attach a photo, then describe the single change you want in plain language instead of generating from scratch. GPT Image 2 uses its native Thinking Mode to reason about your existing image before it edits, applies the change, and keeps everything you didn't mention. Because it runs in a conversation — in ChatGPT or through the API — you can stack edits one at a time: swap the background, then fix the lighting, then remove an object, reviewing each result before the next request. If an image is roughly 80% right, editing it beats starting over.
Why do I have to say what should stay the same?
An edit model can only protect what you name. If you say "replace the background" but never mention the person, GPT Image 2 may subtly re-render their face, pose, or lighting to fit the new scene. Adding "keep the subject's face, pose, hair, and lighting exactly the same; change only the background" locks identity so the one thing you asked for is the only thing that moves. Naming what must stay identical is the single most important habit for clean, believable edits.
How do I keep faces consistent across edits?
GPT Image 2 holds faces and products consistent, and it accepts reference images. Attach the photo you're editing plus a clear reference of the same person, and lock identity in the instruction — "keep the face, bone structure, and skin texture faithful to the reference; this must clearly be the same person, no idealizing." Naming skin texture and telling it not to smooth or reshape the face stops the model from swapping in a prettier stranger. Re-upload the original reference if a later edit starts to drift.
Where can I use these GPT Image 2 editing prompts?
Use them in ChatGPT's image tool, where you attach a photo and type the edit, or through the OpenAI API as gpt-image-2 with your input image and prompt. The API exposes a quality setting (low, medium, high) and output formats (PNG, WebP, JPEG). The prompt wording on this page is identical in both places — paste any block straight into ChatGPT to try it, or wire it into the API for batches and automation.
Can GPT Image 2 remove a background or output a transparent PNG?
Yes. GPT Image 2 outputs PNG and WebP with a true alpha channel. In a prompt, ask for the subject "isolated on a transparent background, PNG with transparent background, no background fill." In the API, set background:"transparent" with output_format png or webp — JPEG silently falls back to opaque. For clean cutouts, keep the subject well-separated from the original background and ask it to preserve fine hair and edge detail.
Can it restore or colorize old photos?
Yes, and it's one of its strongest uses. Attach a scanned photo and ask it to remove scratches, dust, creases, and fading while keeping faces, clothing, and composition exactly the same. For colorizing, tell it to add natural, period-accurate color without altering identity or expression. Because Thinking Mode reasons about the image, it rebuilds damaged areas plausibly rather than smearing — but review results, since it infers detail it can't actually see. Request 2K output for print-grade restorations.