AI product photo editing: what's safe to change and what isn't
Short answer: Use AI to change what's around your product, never the product itself. Background removal, cleaner lighting, consistent cropping and lifestyle scenes for secondary images are fair game; altering the item's color, shape, texture, labels or flaws is not. Keep the main image clean (no added text, badges or logos), keep the AI metadata that editing tools embed, and store every original so you can redo a catalog when a channel's rules change.
AI editing tools have made it possible to turn a phone photo on a kitchen table into something that looks like a studio shot in seconds. That's genuinely useful for small sellers. It also makes it easy to drift from "nicer photo" to "photo of a slightly different product", which is where returns, bad reviews and listing disapprovals come from. Below are the common myths, what's actually true, and a workflow that stays on the right side.
Myth vs reality
Myth 1: "If the AI can make it look better, use it everywhere."
Reality: the main image has the strictest rules. Google's Merchant Center image requirements, as of November 2026, ask for an unobstructed, clean view of the item for sale and warn that post-production text, marketing callouts or graphics cause disapproval. Lifestyle and in-context shots are welcome, but as additional images through separate attributes (additional_image_link, lifestyle_image_link), not as the primary image.
So: clean, accurate main image; creative AI scenes in the supporting slots.
Myth 2: "Removing the background is always safe."
Reality: mostly, yes, and it's the single most useful AI edit for a small catalog. But check three things after every removal:
- Edges: fur, lace, hair, glass and fine chains often come out jagged or partly erased.
- Shadows: a product floating with no shadow can look fake; a soft added shadow often looks more natural.
- Missing parts: straps, cables or transparent elements sometimes vanish with the background.
Zoom in to 100% before publishing.
Myth 3: "A quick 'best seller' badge on the photo helps it stand out."
Reality: on the main image, it's a fast route to disapproval. Google lists calls to action ("Buy", "Shop now"), prices and offers ("Free shipping"), promotional adjectives ("Best", "Cheap"), condition labels ("New"), barcodes and QR codes, watermarks, and logos or store names added in post-production as things to avoid. Borders are out too. Put that messaging in the title, description or ad copy instead.
Myth 4: "Generative fill is fine as long as it looks realistic."
Reality: realism is the problem when it changes the product. Generative fill that extends a backdrop, removes a stray cable from the table, or fills an empty corner is harmless. Generative fill that "fixes" a scuff on a refurbished item, smooths a fabric's real texture, adds a feature, or regenerates a label is changing what you're selling. If a customer would be surprised when the parcel arrives, the edit went too far. Make sure your images honestly represent the product; misleading pictures cause problems well beyond your listings.
Myth 5: "Color correction is just making it look nice."
Reality: color is a product fact. Google asks that variant images accurately reflect the exact color, pattern, finish and material, with one dedicated image per variant rather than one photo reused across colors. Correct white balance so the product matches reality under daylight, then stop. Don't let an auto-enhance push a navy sweater toward royal blue.
Myth 6: "Once it's edited, the metadata doesn't matter."
Reality: Google's requirements state that images created with generative AI must contain metadata indicating that, giving the IPTC DigitalSourceType tag (for example TrainedAlgorithmicMedia) as an example, and ask sellers not to strip those embedded tags. The IPTC vocabulary also includes a value labeled "Edited using Generative AI" (compositeWithTrainedAlgorithmicMedia) for composites. Some generative tools embed these tags themselves (Google names its Product Studio), while image compressors with a "strip metadata" setting can remove them. Check your pipeline doesn't wipe them.
Myth 7: "AI can fake a whole product shot from a description."
Reality: Google's rules say the primary image must clearly display the exact item being sold, and they rule out placeholders, mockups and illustrations that don't depict the physical product (with narrow category exceptions such as some hardware, vehicle parts and software). A generated image of what the product "probably looks like" isn't a product photo.
A catalog workflow that holds up
1. Shoot consistently. Same spot, same light, same angle set for every product: front, back, side, detail, scale. Good input makes AI edits faster and safer.
2. Keep originals untouched. Store the original files in a folder named by SKU before any editing. When a marketplace changes its rules, or you change your site's style, you redo from originals rather than editing edits.
3. Do the safe edits in a batch.
- Background removal or replacement with a plain, consistent backdrop.
- Exposure and white balance correction to match reality.
- Cropping so the product fills a consistent share of the frame. Google's guidance suggests the product take up no less than 75% and no more than 90% of the image.
- Size: Google asks for at least 500 × 500 pixels and recommends around 1500 × 1500 or above, with files under 16 MB.
4. Review against the real item. Put the edited image on screen next to the physical product. Check color, texture, labels, flaws and parts. This step is where most problems are caught.
5. Make lifestyle images separately. If you use AI to place a product in a scene, label them as secondary images in your workflow, keep the product pixels untouched, and use them in the additional image slots, on social and on your own product pages below the main shot.
6. Export without stripping AI metadata. Check your compression plugin or CDN settings.
7. Log what you did. A simple column in your product sheet: "edits: bg removed, WB corrected" or "lifestyle: AI scene". It helps when a channel queries an image, and when a new staff member takes over.
Quick reference
| Edit | Main image | Additional images |
|---|---|---|
| Background removal/replacement | Yes, plain and clean | Yes |
| Lighting and white balance | Yes, toward accuracy | Yes |
| Removing dust or props from the scene | Yes | Yes |
| AI lifestyle scene | No | Yes |
| Text, badges, logos added | No | Check each channel's rules |
| Changing product color, shape, texture or flaws | No | No |
FAQ
Can I use AI-generated backgrounds on my main product image?
A plain, clean background is the safest choice for the main image. Save scenes and props for additional images.
Do I have to label AI-edited photos?
Google's Merchant Center rules require AI-generated images to carry the relevant metadata and ask you not to remove it. Check each marketplace's own rules for any visible labeling.
Is it okay to remove a scratch from a used item's photo?
No. If the scratch is on the item the buyer receives, it belongs in the photo.
What resolution should I export?
For Google listings, at least 500 × 500 pixels, ideally around 1500 × 1500 or more, and under 16 MB per file.
For a repeatable photo pipeline from phone shot to listing that keeps your originals and metadata intact, tell us about your catalog.