Copy-Ready Prompts for Editing Photos in Images 2.6 Without Changing Faces

On November 4, OpenAI updated ChatGPT's image model to Images 2.6. According to OpenAI, images now generate 50% faster, the model follows edit instructions more closely, and you can circle the area you want to change directly on the image. Still, many people get bad results when they edit photos. Ask it to remove a passerby and the whole face changes. Swap the background and the person looks pasted on. A photo-editing prompt has to say what to change, what must stay exactly the same, and how the edit should blend into the original. Below are 24 copy-ready prompts built on that idea: 14 for everyday edits and 10 for turning a selfie into a pro portrait.

The Photo-Editing Prompts

Replace the text in square brackets with your own content. Each prompt follows the order "what to change, what to keep, how to blend." When you edit a person, list the features that must not change first.

Scene	Copy-ready prompt
Full retouch	Retouch this photo: brighten the light, correct the exposure and colors, and make the image look clearer. Keep the person's face, skin tone, hairstyle, body shape and clothes exactly as they are. Keep natural skin texture, and do not smooth it into a plastic look. Only remove obvious small blemishes.
Remove an extra person or object	Remove the [passerby / trash can / power line] from the picture, and fill the area it covered using the surroundings. The filled part must match the nearby ground, wall, light, shadows and perspective. Leave no smeared patches or repeated patterns. Do not change anything else.
Change the background	Change the background to [seaside at dusk / a plain gray studio], and keep the person completely unchanged, including the face, pose, hair and every detail of the clothes. Add shadows and ambient color on the person that match the light direction of the new background, so they look like they were photographed there.
Sharpen	Make this image sharper. Reduce noise, blur and compression artifacts, make the colors more accurate, and give the details crisper edges. Do not change the composition, the person's face, any text, or the position of objects. Only improve the image quality.
Change clothes	Change the person's clothes to [white linen shirt with khaki pants]. Keep the person's face, expression, hairstyle, body shape, pose, hands and background exactly as they are. The new clothes should fit the body, with realistic fabric folds and matching light and shadow.
Extend the canvas	Extend the picture to the [left and right] to a [16:9] frame. Do not change any of the original content. Let the new area grow naturally from the existing environment, light and perspective, so no seam is visible.
Change the time of day	Change this scene to [sunset / night with the lights on]. The sky, light direction, color temperature, shadows and window lights should all change to match, while the positions of the buildings and people stay the same.
Open closed eyes	Change the person's closed or half-closed eyes to natural open eyes. The eye shape, size, gaze direction and lighting must match this face. Keep the expression and facial features the same.
Smooth out wrinkles	Smooth out the messy folds and creases in the clothes. Keep the fabric's natural texture, seams and natural drape, so the clothes look clean and flat but do not look painted on.
Remove text from your own design	Remove the text and logo I circled, and fill the area underneath using the surrounding colors, textures and patterns. Keep the position and colors of the nearby elements unchanged.
Move a product to a new scene	Place this product in a [wooden-table kitchen with natural light]. The product's shape, color, label text and proportions must stay exactly the same. Add contact shadows and reflections that match the new scene.
Match the color across a set	Adjust the colors of this image to match the reference image I uploaded: the same color temperature, contrast and saturation. Keep the image content unchanged and change only the colors.
Change the ID photo background	Change the background to solid [white / blue]. The edges of the hair must be clean and natural, with no fringe and no leftover traces of the original background. The person's facial features, hairstyle and clothes stay unchanged.
Local edit (with circle select)	Only change the area I circled: [replace the cup with a glass one]. Do not change a single pixel outside the circle. The light and perspective of the edited area must match the surroundings.

The last two prompts work with Images 2.6's circle-select tool. First paint over the area you want to change, then send the prompt. A wrong edit then stays much smaller.

10 Prompts That Turn a Selfie into a Pro Portrait

Another common request is turning a clear selfie into a profile photo, a press photo or a magazine-style portrait. Each prompt in this group starts with the same sentence, so the photo still looks like you:

Keep my face shape, facial features, skin tone, hairstyle and signature features unchanged. It must look like me. Keep natural skin texture, and do not over-smooth the skin.

Put this sentence first, then add the scene description from the table below.

Use	Scene description (goes after the sentence above)
Work headshot	Wear a [navy suit with a white shirt], against a plain light gray background, with soft studio light. Look at the camera with a confident, friendly expression. Frame up to the chest.
Personal brand photo	I am a [consultant / creator], sitting in a bright, modern office with a blurred background. My clothes use my brand color [color]. Use natural light from the window. Vertical 4:5.
Magazine cover shoot	Wear a [sharply tailored black suit / simple dress], with a few small accessories. Stand in front of a textured [deep wine red / cream] backdrop. Light from the side, with soft shadows. Half-body framing, vertical 4:5.
Outdoor backlight	At dusk, on a [street / park / beach], with the setting sun lighting the edges of the hair from behind. Blurred background. Wear [clothes]. Relaxed pose, natural smile, as if shot with a portrait lens.
Cafe candid	Sitting by a cafe window, with [a laptop / a coffee] on the table. A natural expression caught while working. Warm daylight, shallow depth of field, vertical framing, suitable for social media.
Executive portrait	Wear a [dark gray] suit jacket, standing in an office with floor-to-ceiling windows. The city is faintly visible outside. Even light, upright posture, calm expression. Do not over-retouch, so it does not look fake.
Creative studio	A plain [blue] backdrop. One colored light in [orange] from one side, neutral light on the face. Clean framing, bold but natural pose, suitable for a personal homepage.
Podcast promo photo	Sitting in front of a professional microphone in a recording booth, wearing headphones. Warm, cinematic lighting, blurred background, and a conversational expression. Leave empty space on the [right] for the show title.
Speaker / author photo	Standing in a [library / event hall], wearing [clothes]. Directional soft light, calm expression. Leave empty space on the [left] for the book title or event details. Vertical 4:5.
Black-and-white close-up	Plain dark background, black-and-white photo, strong contrast, sharp eye detail, with a little film grain. Wear a [black turtleneck]. Close-up, quiet expression, with a photographic look, not an illustration.

What these 10 prompts have in common: the use decides the framing and the empty space (headshots frame up to the chest, press photos leave room for text), the light gets a specific description (for example, window light or backlight), and the expression gets one word, because too many words make the model exaggerate it.

We wrote these scenes and prompts ourselves, adapted from common photo-editing needs. The Images 2.6 update details come from public reporting on the day OpenAI released it. We did not test each prompt side by side.

Why It Works: Three Reasons

• Say what to keep, and the model won't change things by accident. Image models tend to repaint the whole picture. If you only say "remove the passerby," the model may also "improve" the face. Listing what must stay the same sets a boundary for it.
• Describe how to blend, and the edited parts won't show. Light direction and shadows, perspective and texture are exactly where edits most often give themselves away. When you spell these out, the model lines them up.
• Change one thing at a time. If one prompt asks for removing a passerby, a background swap and a color grade all at once, and something goes wrong, you can't tell which step caused it. Split the edits, and each step can be undone and redone.

Use It: A Five-Step SOP

Hand this to your agent (or follow the steps yourself). After each step, check it against the "Done when" column.

Safety Lines

• Only edit images you have the right to use. Removing watermarks or text is only for your own designs and photos. Removing the credit from someone else's work and then using it is infringement.
• Don't change the face or clothes in someone else's photo. Editing another person's appearance without their consent, especially when the result looks like a real photo, may violate their portrait rights and the platform's rules.
• Use retouched ID photos only where they are allowed. Changing the background is fine. Changing facial features or age to get an ID document counts as submitting false material.
• Don't pass a generated portrait off as a real photo. It's fine as a profile picture or for promotion. Do not use a generated image in place of a recent real photo where the platform requires real-identity verification.

Common Pitfalls

• Saying what to change, but not what to keep. The most common result is a face that shifts a little. You may not notice at first, and friends only point it out after you post it: "It doesn't look like you."
• Changing five things in one sentence. When something breaks, you can't tell which step caused it, so you have to start over.
• Using a compressed image as the original. Blur and noise get kept as "content," and the sharpening effect gets weaker too.
• Not zooming in to check. Fingers and earrings go wrong most often, and so do text and hair edges. Shrunk down, you won't see it. Zoomed in, it stands out right away.
• Making a portrait from an unclear selfie. A side-view or backlit selfie leaves the model guessing your features, and the same goes for a selfie with sunglasses on. The result often looks nothing like you. Pick a front-facing photo with even light, and keep the face free of obstructions.
• Assuming clarity can keep improving forever. The model can only guess at details that are missing from the original, so the features and text it fills in may differ from reality.