A client sends you a photo of a vase they want to sell. It was shot on a phone, in a kitchen, under a warm bulb, with a tea towel in the background. You need one clean catalog image before Friday. Reshooting means booking a studio, packing the vase, and hoping nothing chips in transit. The alternative is to fix the photo you were given — but only if you know which problems are fixable and which ones are baked in for good.
Here is how to judge a customer-supplied photo honestly before you promise a client anything. If you want to shortcut the process, the AI image generator on the homepage handles most of the heavy lifting once you know what to feed it.
Start by ruling out the unfixable
Before you touch a single slider, look for motion blur, missed focus, and clipped highlights on the product itself. If the label text is a soft smear because the phone focused on the wall behind it, no amount of processing brings that text back with any accuracy. Same goes for a blown-out highlight sitting right across the product's logo. You can lower brightness; you cannot recover detail that was never recorded.
Crop in to 100% and check the edges of the object against the background. Slight noise is fine. Noise that turns into big coloured blotches when you lift the shadows is not. A quick rule: if you can't clearly read the product's own printed text at full zoom, tell the client you need a re-shoot rather than selling them a fix you can't deliver.
Separate exposure problems from white balance problems
These two get blamed for each other constantly, and fixing the wrong one makes the photo worse. Warm orange casts from indoor bulbs and green casts from fluorescent tubes are white balance. Overall dimness, or a product lit from one side only, is exposure. Open the photo and ask: is the object's true colour wrong, or is the amount of light wrong?
If a white mug looks cream, that's balance. If the same mug looks dark grey with a bright window behind it, that's exposure and possibly a silhouette problem. Correct balance first on a shot with mixed lighting, because fixing exposure on a badly balanced photo will just amplify the cast.
Decide if the background is removal or reconstruction
Simple backgrounds — a plain wall, a bedsheet, a wooden table — come out cleanly. Busy ones with the product touching other objects, or with a reflection of the photographer in a glossy surface, are a different job. The tell is contact: if the object's edge is tangled with another object's edge, expect edge artefacts and plan to zoom in on every boundary before handing anything over.
For catalog work, it's often faster to accept reconstruction: describe the product precisely and generate a fresh clean shot of it rather than fighting a tangled mask. This is where a good generator earns its keep, but only with a description specific enough to matter. "A tall ceramic vase" produces a generic vase. "A tall matte cream ceramic vase, narrow neck, ribbed horizontal bands around the middle, 30cm tall" produces something you can actually use.
Run a two-pass test before committing
Never process the whole set on the strength of one good result. Take the single worst photo in the batch, run your fix, and look at it on a phone screen rather than a calibrated monitor. That's where clients will see it. If the edges look crunchy or the white balance drifts between shots when you compare them side by side, adjust before repeating the process on forty images.
Keep the original untouched in a folder you never overwrite. When a client asks for a warmer version next month, you want to start from the raw file, not from a JPEG that's already been through three rounds of processing.
What to tell the client
Be specific about what you changed. "I cleaned up the background and corrected the colour cast" is honest. "I improved the photo" is going to cause an argument when they ask why the label looks slightly different. If you reconstructed part of the image, say so — most sellers are fine with it as long as the product still looks like their product.
Once you've done this a few times, judging a phone photo takes about ninety seconds, and that judgement is the difference between a smooth job and a week of revisions. The AI image generation tool handles the rendering; your job is deciding whether the source is worth rendering at all.
Batch the work in sets of similar photos
If you have twenty products from the same seller, don't process them one at a time in whatever order they arrived. Group them by lighting condition first — all the window-lit shots together, all the warm bulb shots together. Your correction for one group will hold for most of the rest, and you'll spot the three outliers immediately instead of discovering them after you've already delivered nineteen files.
The same logic applies to backgrounds. Ten items on the same table can share one background treatment. Two items on a different table need their own pass. This kind of sorting takes five minutes and saves an hour of small manual corrections.