Photo Editing Prompt: Describe the Change, Not the Whole Scene

A photo editing prompt works differently from a prompt for generating something from scratch. With generation, you describe the entire frame. With editing, you describe only the change. Try to describe everything anyway, and the model treats your request as an instruction to redraw the whole photo.

The structure of an editing prompt

What to change. One change at a time: the background, a color, the lighting, an object you want gone.

What to keep. A specific list: the subject itself, its shape, any text, the shadows.

How exactly. What the new background should look like, what shade of color, where the light comes from.

What not to touch. Don't add objects, don't shift the composition, leave the edges of the frame alone.

You can arrange these blocks in any order, but skipping the second one is the single most common reason an edit comes out wrong.

Common mistakes with photo editing prompts

Describing the whole scene. List everything in the frame and the model reads it as a request to generate the image fresh.

Stacking several edits into one request. Background, light, and color in a single pass gives you an unpredictable result.

Vague praise words like "make it better" or "cleaner" or "nicer." They don't actually tell the model anything.

Skipping the "don't touch" list. If you only ask for an edit without restrictions, the model assumes everything is fair game.

Being too general about what to preserve. "Keep the product" does less work than "keep the shape, color, and label text on the packaging unchanged."

The order edits should happen in

Start with anything affecting the whole frame: cropping, overall lighting.

Then the background.

Then the object itself: color, details.

Then small fixes: removing dust spots, cleaning up glare.

Finish with resolution and sharpening.

Each step builds on the result of the one before it. That makes it much easier to tell which edit broke what. I've written more about the overall workflow in this piece on photo processing for online listings.

When it's easier to just regenerate

If you've stacked more than three edits and each one pulls along another fix.

If the subject looks noticeably different from the original after two rounds of edits.

If you're trying to fix something that should've been set at the shoot: angle, perspective, where the object sits in the frame.

Editing stays cheaper than regenerating only up to a point. Past that point, you're spending more time patching the image than a fresh generation would have taken.

How to check your work after each edit

Compare against the previous version, not the original. That's where you actually notice what changed.

Check text and small details first. They're the first thing to drift.

Look at the edges of the frame. The model sometimes fills in extra detail there without being asked.

Check the shadows. After a background swap, they often still match the old lighting.

Save your intermediate versions. Being able to roll back one step saves more time than careful wording ever will.

I keep a folder of versions for every edit and still lose track of the right one sometimes. Staying disciplined about this is my weak spot. Numbering files by step has made this noticeably easier.

FAQ

How is an editing prompt different from a generation prompt? An editing prompt describes the change, not the whole scene. Describe the full frame and the model reads it as a request to generate something new.

How many edits should I make at once? One. Multiple changes in a single request give unpredictable results, and you lose any way to tell what actually worked.

What should always be in the prompt? A clear list of what needs to stay the same: shape, color, text, shadows.

What order should edits happen in? Cropping and overall light first, then the background, then the object, then small fixes, then sharpening and resolution last.

Sources

1. Flami: photo processing for online marketplace listings
2. Flami: negative prompts, what to cut
3. Flami: prompts for improving photo quality