October 6, 2026
22 min read
ChatGPT photo editing commands are plain sentences, and there are no secret codes behind them. No slash command, hidden menu, or pasted string edits a photo in ChatGPT. The "code" in every viral video is a short instruction typed under an uploaded photo, and the ones that work share one structure: the change, the region, a keep list, and the light. Here are 30 of them, written tight enough to paste and run.
What Makes ChatGPT Photo Editing Commands Work
Say it plainly first: there are no ChatGPT codes for photo editing. People search for "chatgpt codes for pictures" and "chatgpt / commands for photo editing" because a video showed four lines of text pasted under a selfie and a finished result three seconds later. The four lines were an instruction in English. ChatGPT Images 2.0 reads the uploaded photo and the sentence together, then regenerates the frame to match. Nothing in the sentence is syntax, nothing is case-sensitive, and nothing unlocks a mode. The same words typed with different punctuation do the same thing, and a command written in your own words with the same four parts works exactly as well as a copied one.
What the videos get right is brevity. A command that fits on your screen is one you will paste correctly, and a tight command leaves the model fewer places to improvise. What the videos leave out is the structure that made the result hold together, so the copied "code" works on the creator's photo and falls apart on yours. Every command below carries that structure in full, in as few words as it takes, which is the difference between a command you can reuse on next week's photo and a code that worked once on someone else's.
A command has four parts, in a fixed order. The change ("remove," "replace," "relight," "convert"), the region ("the trash can at the left edge," "the background," "my glasses lenses"), the keep list ("keep my face, pose, clothing, and framing exactly as they are"), and the light ("match the existing light from the upper left"). Drop any one and the model fills the gap with a guess, and the guess is where the odd hand, the shifted jaw, or the wrong shadow comes from.
The region does the targeting. "Fix the background" is a different request from "replace the background with plain light gray and match the shadow direction to the light from the left." The first lets the model decide what a fixed background looks like; the second leaves nothing to decide. Point at the thing by position and name, the way you would direct a person standing at the photo with you.
The keep list is why the video result still looked like the person. Without the selection brush, ChatGPT regenerates the whole frame from your photo plus your words, so a face, a hand, or the crop can drift while the model works on something else. A keep list that names face shape, features, skin tone, hair, expression, pose, framing, and proportions tells the model what to notice and leave alone. "Keep everything else the same" is too loose to hold.
Light is the part people skip, and the part viewers notice. A new background lit from the right under a face lit from the left looks pasted on. State where the existing light comes from and ask the fill, backdrop, or added object to agree with it. One phrase, every time, and name the direction as it appears in your photo rather than copying the direction from a video.
Short works when every word has a job. The commands below run three to six lines. They are shorter than the templates in the hub post because they drop the optional detail and keep the four parts. If a result drifts, add detail to the keep list first, then to the region, and leave the change itself alone.
Brush when the edit is local. After uploading, open the photo and paint the exact region with the selection brush, then paste the command. Only the painted area is regenerated and the face is never inside it. Use the brush for an object, a blemish, a sign, or a shirt; describe without brushing for a relight, a restyle, or a background swap that touches hair edges.
Info
ChatGPT does have real slash-style shortcuts. Typing a forward slash in the message box opens a short menu of tool shortcuts, such as web search, deep research, canvas, thinking longer, and image creation. Each one switches a tool on. None of them contains an editing instruction, and none of them edits an uploaded photo. The image shortcut starts a generation from text; the words you type after it still do all the work. Any video or post that shows a "secret / command for photo editing" is showing a tool switch followed by an ordinary sentence.
Every command here is a compressed version of a template in the 40 ChatGPT photo editing prompts hub, where each edit comes with a longer keep list and more variations. When you need a command for an edit that is not here, describe the photo and the change to the AI prompt generator and it writes the four parts for you.
30
How to use a ChatGPT photo editing command, start to finish:
1
Upload the photo in a new chat so earlier images cannot bleed into the result, and check it displays full size and the right way up.
2
Copy one command, paste it under the photo, and replace every [BRACKETED] value with what is in your photo: the object, the edge, the color, the light direction.
3
If the command is one of the short ones and your photo has a face, paste the Face Lock Prefix (command 23) in front of it.
4
For a local edit, open the photo, paint the exact region with the selection brush, then send. For a global edit, send without brushing.
5
View the result at 100% and compare eyes, teeth, hands, hair edges, and any text against the original.
6
If it is close, resend with one change, such as the light direction. If the face drifted, start a new chat with the original photo and a longer keep list.
Warning
Use these commands on photos you own or have permission to edit. Do not remove, add, restyle, or relocate people without their consent, and do not paste a "watermark removal code" on someone else's work: removing a credit or watermark you do not own is a rights violation whatever the prompt is called. Photos of children, ID documents, and medical images deserve extra care. Delete the chat when you are done.
Quick Fixes (1–6)
These are the ChatGPT photo editing commands people paste first: remove the thing, straighten the frame, sharpen the soft shot, brighten the dark one. Brush the object in commands 1 and 6; the rest touch the whole frame, so describe and rely on the keep list. "Only" and "exactly" are doing real work in these, so do not trim them.
1. Remove an Object
code
Remove the [TRASH CAN] at the [LEFT EDGE] and continue the
[PAVEMENT AND WALL] behind it with matching grain and perspective.
Match the existing light from the [RIGHT]. Keep my face, pose,
clothing, framing, and proportions exactly as they are. No blur
patch, no repeated texture.What you'll get: The object gone and the background continued without a smear.
Variation: Brush the object first and keep only the fill and keep lines.
2. Straighten and Crop
code
Level the [HORIZON] and crop to [4:5] with my full head and
shoulders in frame. Fill empty corners with matching background
only. Do not stretch, warp, or regenerate anything. Keep my face,
colors, and lighting exactly as they are.What you'll get: A level frame at the ratio you asked for, no warped edges.
Variation: Swap 4:5 for 1:1, 3:1, or 1:3; ChatGPT Images 2.0 supports the full range.
3. Sharpen a Soft Shot
code
Sharpen this slightly soft photo: eyes, hair edges, and fabric
only, no halos, no invented detail. Keep my features, expression,
skin texture, pose, and framing exactly as they are. Same colors
and light. Output at 2K, same aspect ratio.What you'll get: A crisper photo that still looks like a photo.
Variation: For motion blur, add "the blur is directional, left to right."
4. Brighten a Dark Photo
code
Brighten this underexposed photo by about [1.5] stops. Lift the
shadows on my face, protect the [WINDOW] highlights from clipping,
keep skin and fabric texture. Keep my face, pose, framing, and the
room exactly as they are. No HDR glow, no warm cast.What you'll get: A readable exposure with the highlights intact.
Variation: Add "keep the mood dim, just make faces readable" for evening shots.
5. Reduce Noise
code
Reduce the grain in this low-light photo while keeping real
texture in skin, hair, and fabric. Do not smooth my face or
sharpen edges. Keep the colors, the mood, my features, pose, and
framing exactly as they are. No plastic look.What you'll get: Less grain, same texture, no waxy skin.
Variation: Add "leave a little fine grain so it still reads as film."
6. Declutter the Background
code
Remove the clutter on the [DESK AND SHELF] behind me and leave the
furniture bare, continuing each surface with its own texture and
the room's light from the [WINDOW LEFT]. Keep me, my pose,
clothing, and framing exactly as they are. No new objects.What you'll get: The same room with the mess gone and nothing invented.
Variation: Name what stays: "keep the plant and the lamp, remove everything else."
Background Commands (7–11)
Background commands fail when the model treats "new background" as permission to redraw the subject, so the keep list names hair edges, glasses, and shadow direction every time. Describe these rather than brushing them; a brush stroke around hair leaves fringe. For backdrops generated from scratch, with no photo attached, see ChatGPT background prompts.
7. Blur the Background
code
Blur the background to a soft f/1.8 look so I stand out. Keep my
face, hair edges, glasses, clothing, pose, and framing exactly as
they are and fully sharp. Blur increases with distance. No halo
around my hair, no blur on my shoulders.What you'll get: A portrait-lens look with a clean edge around you.
Variation: Ask for "a light f/4 blur" to keep the setting readable.
8. Plain Color Backdrop
code
Replace the background with a plain, even [LIGHT GRAY #d9d9d9]
seamless. Match the shadow direction to the existing light from
the [UPPER LEFT]. Keep my face, hair edges, glasses, clothing,
pose, and framing exactly as they are. No halo, no cut-out look.What you'll get: A studio-style backdrop that agrees with your light.
Variation: Use #f2f2f2 for high-key, #2b2b2b for dark, or a brand hex code.
9. Outdoor Swap
code
Put me in front of [SOFT GREEN TREES, OUT OF FOCUS] in overcast
daylight. Keep the light on my face soft and neutral to match.
Keep my face, expression, hair, clothing, pose, and framing
exactly as they are. No sharp leaves, no fake bokeh, same aspect
ratio.What you'll get: An outdoor portrait where the light and the scene agree.
Variation: Swap the trees for "a quiet beach at overcast midday."
10. Transparent Cut-Out
code
Cut me out and export a PNG with a true transparent alpha channel,
not a drawn checkerboard. Clean hair edges, no background fringe.
Keep my face, hair, and clothing exactly as they are. No shadow,
no outline, same proportions.What you'll get: A cut-out you can drop on any color; test it on a dark layer.
Variation: Add "a white sticker outline about 4% of the width" for stickers.
11. Extend the Background
code
Extend this photo to [16:9] by continuing the [WALL AND FLOOR]
outward on both sides with matching texture, perspective, and the
light from the [LEFT]. Do not scale or move me. Keep my face,
pose, clothing, and the original pixels exactly as they are. No
new objects.What you'll get: A wider frame with the original untouched in the middle.
Variation: Extend upward only for a 4:5 post from a landscape shot.
Light and Color Commands (12–16)
These are global edits: describe, do not brush. Name the light source, its direction, and its color, and say what skin should look like afterward, because "warmer" on its own turns faces orange. For a longer set of exposure, color, and sharpness fixes, the sibling post on ChatGPT photo enhance prompts goes deeper than these five.
12. Golden Hour Relight
code
Relight for golden hour: low warm sun from [CAMERA LEFT, BEHIND
ME], soft warm rim on my hair, warm fill on the shadow side. Keep
my features, expression, pose, clothing, and framing exactly as
they are. Skin warm and realistic, no orange cast, no flare.What you'll get: Evening light on the same photo, face still yours.
Variation: Put the sun "directly behind me" for a strong rim and a darker face.
13. Fix White Balance
code
Correct the white balance: the [LAMP] side of my face is orange
and the [WINDOW] side is blue. Make whites read white and both
sides of my face match with natural skin. Keep exposure, features,
pose, and framing exactly as they are. No smoothing.What you'll get: Neutral whites and one skin color across the whole face.
Variation: Brush only the off side and ask it to "match the rest."
14. Black and White
code
Convert to black and white with deep blacks, detailed highlights,
and smooth skin midtones, red-filter effect. Keep the composition,
my face, pose, and clothing exactly as they are. No sepia, no
vignette, no selective color.What you'll get: Real tonal depth instead of a flat gray desaturation.
Variation: Swap "red-filter" for "green-filter" to deepen lips and darken skin less.
15. Richer Color Without Orange Skin
code
Increase color saturation in the [SKY, GRASS, AND CLOTHING] by
about 20% and leave skin tones exactly as they are. Keep the
exposure, light direction, my features, pose, and framing
unchanged. No orange skin, no neon greens, no HDR look.What you'll get: Punchier scenery with your skin the color it was.
Variation: Name one color to boost: "only the blue of the water."
16. Fill the Shadow Side
code
Add soft fill light to the shadowed [RIGHT] side of my face as if
from a reflector at camera right, matching the existing key light
from the [LEFT]. Keep my face shape, features, skin tone,
expression, pose, and framing exactly as they are. No flat look,
no new catchlights, no smoothing.What you'll get: A readable shadow side that still has shape.
Variation: Ask for "a small warm bounce" to keep some contrast.
Style and Filter Commands (17–22)
Filter-style commands are what most "chatgpt image codes" videos show, and they change every pixel, so likeness is the whole game. Describe the look rather than naming an artist or a studio; ChatGPT often declines named styles and a described look matches closer anyway. Keep face shape, hairstyle, glasses, expression, and outfit colors in every keep list.
17. Film Grain Look
code
Give this photo a warm color-negative film look: soft contrast,
lifted blacks, fine grain, pastel highlights. Keep the
composition, my face, expression, pose, and clothing exactly as
they are. No light leaks, borders, dust, or date stamps.What you'll get: A film grade that leaves the photo itself alone.
Variation: Upload a reference photo and say "match the first to the second's grade."
18. Cinematic Teal and Orange
code
Grade this photo cinematic: cool teal shadows, warm highlights,
gentle contrast, widescreen feel without black bars. Skin stays
natural, never orange. Keep my face, pose, clothing, and
framing exactly as they are. No vignette, no grain.What you'll get: The movie-poster grade with normal skin.
Variation: Swap teal for "deep green shadows" for a thriller look.
19. Watercolor
code
Turn this photo into a loose watercolor on textured paper: soft
wet edges, white paper in the highlights, [FIVE]-color palette.
Keep my likeness: same face shape, hairstyle, glasses, expression,
pose, and outfit colors. No outlines, no signature.What you'll get: A painting that is still recognizably you.
Variation: Swap "loose" for "tight botanical-illustration precision."
20. Animated-Film Character
code
Restyle as a 3D animated-film character: soft skin shading,
simplified hair clumps, slightly larger eyes. Keep my face shape,
skin tone, hairstyle, glasses, expression, pose, and outfit colors
recognizable. Soft key light from the [UPPER LEFT]. No text, no
extra characters.What you'll get: A stylized character that friends recognize at a glance.
Variation: Ask for "a flat 2D vector illustration with bold outlines."
21. Oil Painting
code
Repaint this photo as an oil painting: visible brush strokes,
thick highlights, canvas texture, warm palette. Keep my face
shape, features, hairstyle, expression, pose, and framing
recognizable and the composition unchanged. No frame, no
signature.What you'll get: Brushwork with the composition and likeness intact.
Variation: Ask for "impasto only in the highlights, smooth elsewhere."
22. Faded Vintage Print
code
Make this look like a faded [1970s] print: warm color shift,
lower contrast, soft grain, slightly washed blacks. Keep my face,
expression, pose, clothing, and framing exactly as they are. No
scratches, no borders, no date stamp.What you'll get: An album-print feel without fake damage.
Variation: Add "a thin white border like a printed snapshot."
Face-Safe Commands (23–26)
The complaint under every "chatgpt photo code" video is the same: the background changed and so did the face. These four ChatGPT photo editing commands exist to stop that. Command 23 is a prefix you paste in front of any other command; 24 to 26 are the edits where the face is closest to the region being changed. None of them alters face shape, age, or skin tone, and you should not ask for edits that do.
23. Face Lock Prefix
code
Before any change: keep my face shape, eye shape and color, nose,
lips, teeth, skin tone and texture, hairline, hair, expression,
pose, framing, and body proportions exactly as in the uploaded
photo. Do not retouch, slim, smooth, or brighten my face. Now:
[YOUR COMMAND].What you'll get: Any other command on this list with the likeness pinned down.
Variation: Add specifics the model should notice: "the gap in my front teeth, the mole on my left cheek."
24. Light Retouch Without Smoothing
code
Light retouch only: remove the [BLEMISH ON MY CHIN] and stray
flyaway hairs. Keep pores, freckles, and real skin texture. Do not
smooth, reshape, brighten, or whiten anything. Pose, framing, and
background stay exactly as they are. No airbrushed look.What you'll get: A good-skin-day version of the same photo.
Variation: Add "reduce the shine on my forehead by half."
25. Fix Glasses Glare and Red-Eye
code
Remove the reflections on my glasses lenses and fix the red-eye
so my eyes show their natural [BROWN] color. Keep the frames,
my eye shape, gaze, face, and expression exactly as they are.
Match the light on the rest of my face. No redrawn eyes.What you'll get: Clear lenses and natural eyes behind them.
Variation: Brush the lenses only and drop the red-eye line if that is the sole issue.
26. Group Photo, Every Face Kept
code
Remove the [EXIT SIGN AND CABLES] above the group and continue
the [WALL] with matching texture and light. Keep every person's
face, expression, pose, clothing, and position exactly as they
are, nobody added or removed. No smoothing on anyone.What you'll get: The distraction gone and the whole group unchanged.
Variation: Brush the object so no face is inside the regenerated region.
Text and Social Commands (27–30)
Text in images is readable and reliable in ChatGPT Images 2.0, which is why the caption and watermark "codes" spread fastest. Put the exact string in quotes, say "spelled exactly as written," and forbid extra words. The crop commands here extend the background rather than stretching it, so you stay the same shape at every ratio, and the watermark command starts with a rights statement because that is the one edit ChatGPT will ask about.
27. Caption Band
code
Add a solid [#222222] band across the bottom 12% with the text
"[YOUR CAPTION]" in clean white sans-serif, left-aligned, spelled
exactly as written, no extra words. Keep the photo above the band
unchanged: same crop, colors, and subject. Output in 4:5.What you'll get: Correctly spelled text with the photo untouched above it.
Variation: Move the band to the top and ask for two lines.
28. Story Crop
code
Reframe this photo to 9:16 for a story. Keep me at the original
scale and position; extend the [BACKGROUND] above and below with
matching texture and light from the [LEFT]. Keep my face, pose,
and clothing exactly as they are. No stretching, no new objects.What you'll get: A vertical frame with the original photo intact inside it.
Variation: Ask for 4:5 for a feed post or 1:1 for a profile grid.
29. Remove My Own Watermark
code
This is my own photo and I own the rights. Remove the
[SEMI-TRANSPARENT LOGO IN THE CORNER] and fill the area with the
matching [SKY] behind it, same grain and softness. Keep every
face, object, color, and the framing exactly as they are. No
blur patch.What you'll get: Your old overlay gone with no smudge left behind.
Variation: Brush the mark and keep only the fill and keep lines.
30. Add a Small Brand Mark
code
Add "[YOUR NAME]" as a small text mark in the [LOWER RIGHT]
corner: clean sans-serif, white at about 40% opacity, under 10%
of the image width, spelled exactly. Keep the photo itself
unchanged: same crop, colors, light, and subject. No icon, no
shadow.What you'll get: A subtle mark you can repeat across a set.
Variation: Upload your logo as a second image and say "place this exact logo."
Tips for Better Results With ChatGPT Photo Editing Commands
Expect a different result from the video, every time. The creator's photo had its own light, background, and resolution, and the creator probably ran the command several times before recording the one that worked. ChatGPT Images 2.0 has no fixed seed you can paste, so the same words on a different photo produce a different edit. Treat a command as a starting point and tune one variable per retry: the light direction first, then the fill description, then the keep list.
Paste the Face Lock Prefix before any command that worries you. It costs one line and removes the most common complaint. If the face still drifts after two attempts in the same chat, stop editing the drifted version: open a new chat, upload the original, and run the prefix plus the command once.
One command per message. "Remove the sign, blur the background, and make it golden hour" gives the model three regions and three lights to reconcile, and the keep list protects nothing well. Run three commands in sequence instead, checking the face after each.
Fill every bracket with what is in your photo. [LEFT EDGE] is a placeholder for your left edge, and [UPPER LEFT] is where your light comes from, not the video's. A bracket left unfilled is the fastest way to get a confident, wrong edit.
Pick the tool for the edit. For a hard local edit where nothing else may move, Gemini's Nano Banana is the stronger tool, and the best Nano Banana prompts collection covers it; for relighting, restyles, and anything with text, start in ChatGPT.
Tip
Instant mode runs these commands on every plan, including free, where a small daily allowance that OpenAI adjusts over time makes each attempt count. Paid plans add Thinking mode, which plans the layout before drawing, checks its own output, and can return up to eight coherent images from one prompt, useful when you want a command run across several photos at once.
Generate Custom ChatGPT Photo Editing Commands
Your photo has its own light, its own clutter, and its own keep list. Describe the edit in plain English, including what the light is doing and what must stay fixed, to the AI prompt generator and it writes the four parts, keep list included; the image prompt generator is tuned for visual work, and the Template Builder saves a face-lock block you can paste in front of any command.
For the long-form version of every command here, with more variations and the reasoning behind each keep list, go back to the 40 ChatGPT photo editing prompts hub. For a selfie headed to a profile, ChatGPT profile picture prompts composes for the circle crop and locks the likeness the same way command 23 does.
Four parts, filled brackets, one command per message, face checked at 100%.