The ChatGPT Photo Editor: How to Edit Step by Step
Aiko Tanaka
The ChatGPT photo editor isn't a separate app. It's the image side of ChatGPT itself: you upload a photo, say what to change, and ChatGPT Images 2.5 hands back an edited version. Since the 2.5 update on September 8, 2026, you can also pin a comment to an exact spot on the image, which makes targeted edits much easier to ask for.
Short answer. To edit a photo with ChatGPT, open a new chat, upload the photo with the plus button, and type one clear instruction such as "Remove the person on the right; keep everything else the same." ChatGPT regenerates the photo with the change. Refine with follow-ups or in-image comments, then download. It handles content edits well, but every edit redraws the whole image, so exact crops, sizes and color values need a canvas editor.
How to Edit a Photo in ChatGPT, Step by Step
The flow is the same on chatgpt.com, the desktop app and the mobile app, and it works on the free plan.
- Start a new chat. Old instructions in a long thread can bleed into the edit.
- Upload the photo. Tap or click the plus button next to the message box and choose the image. Use the largest version you have; detail lost in a small upload doesn't come back.
- Write one instruction. Name the change, where it is, and what must not change: "Change the wall behind her to sage green. Keep her face, hair and clothing exactly the same."
- Send and compare. Look at the edited area first, then check the parts you didn't ask to change: faces, hands, text, logos and edges.
- Refine with a comment or a follow-up. In Images 2.5, click a spot on the image and leave a comment like "remove this reflection" to target that area. Or reply in the chat: "Good. Now make the light slightly warmer."
- Download the version you want. Open the image and use the download button, or save to Photos on mobile. Earlier versions stay in the thread if a later edit went too far.
When an edit keeps missing, it usually works better to go back to the original photo in a new message than to keep correcting the latest version. Each round is generated from the image before it, so small errors add up.
Edit Instructions That Work in ChatGPT
The instructions that come back clean have the same shape: one change, a location, and a "keep" clause. Ending with "keep everything else the same" does more than any other phrase to protect the parts of the photo you didn't mention.
| Edit | Instruction that works |
|---|---|
| Remove an object | "Remove the trash can on the left and fill the space with the brick wall behind it." |
| Replace a background | "Replace the background with a plain light-gray studio backdrop. Keep the person exactly as is." |
| Change a color | "Change the sofa to deep olive green. Keep the fabric texture and shadows." |
| Fix the light | "Make the lighting look like soft late-afternoon sun from the left. Don't change the people." |
| Change the season | "Make this a snowy winter scene. Keep the house and car unchanged." |
| Add text | Add the words "Open Saturday" in white bold sans-serif at the top center. |
| Restyle | "Turn this into a watercolor illustration, same composition." |
| Clean up a portrait | "Remove the stray hairs across her forehead. Keep her face, skin texture and expression identical." |
What tends to fail: stacking three edits in one message, vague words like "better" or "pop", and anything with a number in it ("brighten by 15%", "crop to 1080 × 1350"). ChatGPT interprets numbers loosely because it isn't moving a slider, it's drawing a new picture that it judges to be "15% brighter".
Where the ChatGPT Photo Editor Breaks
ChatGPT doesn't change the pixels you uploaded. GPT Image 2.5 looks at your photo and your instruction and generates a new image meant to satisfy both. That's why it can do things a classic editor can't, like change the season or invent a background that was never there. It's also why some simple edits are unreliable.
Everything is redrawn. Even a one-word change regenerates the whole frame. Parts you didn't mention are re-created rather than preserved, so they can shift.
Drift adds up. Images 2.5 is better at this than earlier versions. Implicator's report on the launch says the update preserves subjects from reference photos more reliably, keeps earlier changes more consistent across repeated edits, and keeps image quality from degrading over time. Better isn't zero, though. After five or six rounds, faces, product labels and fine patterns often look subtly different from the original.
No exact crop or size. "Make it square" produces a recomposed square image, not your photo with the sides trimmed. There's no crop box, no straighten tool and no way to enter 1080 × 1350.
No exact color or exposure. There are no brightness, contrast, saturation, temperature or tint controls, and you can't type a HEX value and expect it back precisely. "Slightly warmer" means whatever the model decides.
Resolution is the model's, not yours. The output size is set by the generation, not by your original file, so a high-resolution upload can come back smaller than it went in.
For a lot of photos none of this matters. A social post that needs a cleaner background, or a mock-up for a meeting, is exactly what ChatGPT does well. It matters when the photo is a product image, a portrait someone will compare against the original, or an asset that has to fit an exact slot.
Photoshop in ChatGPT: Adobe's Option Inside the Chat
ChatGPT also has a second photo editor built in, and it works differently. Adobe connected Photoshop, Adobe Express and Acrobat to ChatGPT as free apps on December 10, 2025. According to Adobe's announcement, Photoshop in ChatGPT can adjust a specific part of an image, fine-tune brightness, contrast and exposure, and apply creative effects such as Glitch and Glow. A March 2026 update added generative edits: Adobe's guide to generative AI in Photoshop for ChatGPT lists adding, removing or replacing elements, replacing and generating backgrounds, and editing specific objects or people, with free daily AI edits for any ChatGPT user.
To use it:
- In ChatGPT, open Settings → Apps & Connectors (Adobe's walkthrough shows each step), choose Adobe Photoshop and click Connect.
- In a chat, upload the photo, click the + button, choose More and pick Adobe Photoshop.
- Describe the edit in plain language, such as "make the people stand out" or "give the sky a grainy retro look".
- Fine-tune the result with the sliders Photoshop returns, then download.
The difference matters for the problem above. Photoshop's adjustments come back with sliders you can move, so tonal and color changes are applied to your photo rather than redrawn from scratch. Its generative edits, like ChatGPT Images, create new pixels, so the redraw caveat applies to those too. For brightness, contrast, color grading and stylized effects, the Photoshop app is the better choice inside ChatGPT; for content changes like "put this product on a marble counter", ChatGPT Images is.
How to Make Precise Edits on a Canvas Instead
The fix for the edits above is to stop regenerating and edit the actual image. Playyy opens any photo, uploaded or generated, on an editable canvas, where each kind of edit has its own control and leaves the rest of the pixels alone.
Crop and straighten. The image cropper has free ratios and social presets, plus a straighten control for tilted horizons. The crop is a trim of your real pixels, so nothing in the frame gets redrawn.
Brightness, contrast and color. The brightness, contrast and saturation editor gives you sliders for brightness, contrast, highlights, shadows, saturation, temperature and tint. You can move warmth up a few points and know that's the only thing that changed.
Recolor one color. The color changer picks up the dominant colors in the photo and swaps one for another, which is the reliable version of "make the sofa olive green". Film-style looks are in the photo filters.
Erase and remove. The object remover erases a selected item and fills the gap, without touching the face two inches away. A one-click background remover cuts out the subject for a clean or transparent background.
Change one element. For the content edits ChatGPT is good at, there's a targeted version: click an element and describe the edit, or use inpaint to replace a selected area. Only the selection is regenerated, so the rest of the photo stays pixel-identical.
The generator in Playyy runs GPT Image 2.5 by default, the same model as ChatGPT Images 2.5. If you'd rather create the photo from a prompt and then fine-tune it, start with the GPT Image 2.5 generator and set the aspect ratio (nine options from 1:1 to 9:16) and resolution (1K, 2K or 4K) before you generate.
If the edited photos are for a brand, the color problem above has a second half: ChatGPT doesn't know your brand, so you re-describe your palette and logo in every chat. In Playyy you paste your website URL once and it builds a brand kit, extracting your brand colors, fonts, voice and tone, and logo, then drafts on-brand designs from it. The finished photo can go into one of 1,000+ templates, and the Prompt Gallery has 1,000+ ready-made GPT Image and Nano Banana prompts when you want a new scene rather than an edit.
Which Edits to Do in ChatGPT and Which on a Canvas
| Edit | ChatGPT Images 2.5 | Canvas editor |
|---|---|---|
| Change the scene, season or style | Strong | Use inpaint or edit elements |
| Remove a large object | Good, may shift nearby areas | Object remover, rest untouched |
| Swap the background | Good | Background remover, exact edge |
| Crop to an exact ratio or pixel size | Recomposes instead | Crop presets and straighten |
| Brightness, contrast, temperature | Approximate | Sliders, exact |
| Recolor to a specific color | Approximate | Dominant-color recolor |
| Keep a face or label identical | Not guaranteed | Untouched outside the edit |
A workflow that uses both: do the creative change in ChatGPT, where describing it is faster, then bring the result into a canvas for the crop, the color match and any clean-up. Doing it in that order means the precise steps happen last, and no later regeneration can undo them.
If the photo started life in ChatGPT, our guide to using the ChatGPT image generator covers the generation side, including the free plan and what Images 2.5 changed. For a deeper look at element-by-element editing of AI images, see the ChatGPT alternative for image editing.
The Bottom Line
ChatGPT works as a photo editor when the change is about content: remove this, replace that, make it winter. Upload the photo, give one instruction with a "keep everything else the same" clause, and refine with in-image comments. It struggles with exact edits, because every change regenerates the whole image: crops, pixel sizes, exposure values and colors that have to match.
For those, edit the photo itself. Open it in Playyy's image cropper or any of the canvas tools above, and make the change once, exactly.
Aiko Tanaka
I work with boutique brands, cafes, creators and small businesses on visual systems, layout discipline, typography and moodboards. My focus is on fast concept exploration that still has a strong design point of view.
Frequently asked questions
Yes. Upload a photo to ChatGPT and describe the change — remove an object, swap the background, change the season, restyle the lighting — and ChatGPT Images 2.5 returns an edited version. It works on the free plan within a daily image limit. Every edit is a regeneration of the whole picture, so it's strongest for content changes and weakest for precise adjustments.
