In six situations, automated retouching is not a slightly worse option but the wrong kind of tool. Most of them are about accuracy and consent rather than looks.
Avoid automated and generative retouching when the photograph is a record someone can check against reality, when the subject's identity is the point, when nobody asked for the change, when the file is covered by an NDA or a release, when the frame will be seen at full size, and when a set has to match. In most other cases it is a useful tool.
This is not an argument against automated retouching tools, which are useful. It is a list of cases where the risk is not slightly worse output but output that causes a problem.
1. The photograph is a record
Product photography, where the customer receives the object. Property listings. Documentation. Anything that may be used as evidence. Medical or dental before-and-after images.
Generative tools add plausible detail, and in a record, added detail is a false statement. A generatively "cleaned" product shot can misrepresent what arrives in the box, which is a commercial and sometimes legal problem rather than a matter of taste.
The rule: if someone can compare the photograph with the real thing, nothing in it may be invented. What commercial retouching can safely change is covered in product and commercial retouching.
2. The subject's identity is the point
Automatic portrait tools pull faces toward an average. The features they remove first are the ones that make a face specific: freckles, moles, asymmetry, scars, lines.
On a portrait commission, a family photograph or a headshot, that is not retouching but replacement. People notice even when they cannot say what changed.
It matters even more when a small face has been enlarged. Rebuilt features are plausible rather than accurate, and a plausible face is not necessarily the same person.
3. Nobody asked for the change
Automatic tools apply their full treatment by default: slimming, smoothing, eye enlargement, teeth whitening, all at once, with nobody deciding.
Changing someone's body or face without being asked is a decision made about them rather than for them. When you ask, the answer is often "leave it".
4. The file is under an NDA or a signed release
Cloud tools upload your file to someone else's servers, and some free ones keep uploads or reserve rights over them.
For a client's portrait, a product under embargo, or anything covered by a release, that is a contractual question rather than a preference. It is not a reason to avoid the tools altogether. It is a reason to decide once which files may go where.
5. The frame will be seen at full size
Automated skin work tends to look fine at 50% and give itself away at 100%. A frame going to print, a full-screen crop or a model's portfolio will be seen at a size where missing pore structure is obvious.
6. A set has to match
Automatic tools work one image at a time. Put forty frames through one and each gets its own reading of what it needed.
The result is forty individually plausible frames that do not belong together, and people notice that across a spread before they notice anything about a single image.
Where automation is fine
Masking. Denoise. Upscaling within sensible limits. Dust and sensor-spot removal. A first culling pass. Adjustments on genuinely uniform sets. Personal work, rough previews, and anything you will review before it goes anywhere.
What these have in common: the output is easy to check, and you are going to check it. AI vs human retouching sets out the full division.
A test before you run the tool
Ask one question before putting a client's frame through an automated tool: would I be comfortable telling the client exactly what was done to this image?
If yes, go ahead. If the frame involves someone's face and the honest answer is "I would rather not go into it", do the work by hand or leave it alone. That is also why our Beauty Retouching keeps real skin texture rather than delivering completely smooth skin.
Be open about what was done
Where an edit goes beyond ordinary cleanup, say so in the delivery note: a generatively extended background, a head taken from another frame in a group shot, a product surface rebuilt after damage. One line is enough, and it means the client never finds out later from someone else.
Consider adding a line to your client agreement about how automated and cloud-based tools are used on their files. It answers the question before it is asked.
And when a client asks what was done to a photograph, answer specifically. "Standard retouching" is not an answer to someone who thinks their face has changed.
Common questions
Is generative fill ever the wrong choice on client work? Yes, when someone can compare the image with the real thing: a product, packaging, a room or a document. Generated detail is plausible, and plausible is the wrong answer when it can be checked.
What is the problem with automatic portrait tools? They pull faces toward an average, and the features they remove first are the ones that make a face recognisable.
Do I need permission to change someone's appearance? Asking is the right habit. Slimming, smoothing and reshaping are decisions about a person, and the answer is often "leave it".
Can I use cloud AI tools on NDA work? Read the tool's terms and your contract first. Some services keep uploads or reserve rights over them, which makes this a contractual question rather than a preference.
Is AI upscaling safe for faces? Be careful. Enlarged faces are rebuilt with plausible features, not recovered ones, so avoid it wherever the person has to be recognisable.
Related reading: 10 Retouching Mistakes Clients Notice Immediately and Best AI Photo Editors in 2026.
BeEdits Editorial Team
Post-production studio
A photography post-production company: high-end retouching, creative editing and scalable production support for photographers, studios and brands.
