Top 5 Mistakes People Make When Using AI Photo Apps
Five errors that account for most disappointing results, why each one happens, and the specific habit that fixes it.
These five account for the overwhelming majority of disappointing results. None is about choosing the wrong app.
1. Asking for a judgement instead of a change
The most common and the most consequential.
"Make this look professional." "Perfect skin." "Make me look better." These feel like instructions and they are not — they are opinions, and a model resolves an opinion by moving toward whatever it has seen labelled that way.
For portraits, that average is younger, more symmetrical and less like you. The drift is not a bug; it is the model doing exactly what was asked, by the only means available to it.
The fix: describe a visible change. "Replace the background with plain light grey." "Even out the lighting on the face, keep the skin texture." "Remove the bin on the left and its shadow." Every one of those is something you could point at.
2. Changing several things at once
"Replace the background and make me look professional and sharpen it" produces a result you cannot diagnose. Something is wrong; you have no idea which clause did it.
Worse, a broad clause bundled with a narrow one runs unopposed. You asked for a background and also, incidentally, gave the model permission to change your face.
The fix: one change per run. Check. Then the next. It costs one extra generation and converts guesswork into diagnosis.
3. Starting from a photo that cannot support the edit
The one people most resist, because it means going back rather than forward.
These tools propagate and refine what is in the image. They do not recover what was never captured. A blurred face has no structure to sharpen, so upscaling produces a plausible stranger. A photo with no separation between subject and background makes background replacement leave a halo in the hair.
The fix: retake it. Five minutes by a window, camera at eye level, at least 1.5 metres away, twenty frames. That beats an hour of editing a poor photo, and it is not close. The method is in getting a studio-quality photo without a studio.
4. Trusting the result instead of checking it
Generative tools fail confidently. They do not produce a visibly broken image and signal uncertainty; they produce a crisp, plausible one whether or not the prediction was right.
Which means your impression of the result is not evidence. And impressions are biased toward whichever image you saw most recently, which is always the new one.
The fix: open the original and the result side by side at full size, and compare specific features rather than the overall effect. For a face: eye spacing, nose bridge, hairline, jaw, skin tone. For anything: the hair-to-background boundary, glasses frames, teeth, any text.
Keep the original. Generative edits have no history to step back through, so the original file is the only comparison you will ever have.
5. Judging at the wrong size
You edit on a large screen at full resolution. The image is displayed at 40 pixels in a feed, or recompressed into a video stream, or printed.
Almost every edit that impresses at full size is invisible at display size, and several — sharpening, saturation — actively degrade after platform compression.
The fix: check at the size it will actually be seen. Shrink to thumbnail. Look at it on a phone. If it does not work there, the time spent at full size was spent on nothing. The platform-specific versions are in Instagram and TikTok guides.
The honourable mention: regenerating instead of diagnosing
Not quite a top-five mistake because it wastes time rather than producing a bad result, but it is what people do instead of the five fixes above.
Generation is stochastic, so a second attempt genuinely does produce a different image, and sometimes a better one. That makes regenerating feel productive. It is the right move roughly twice — after that, the variation between runs is smaller than the problem you are hitting, and you are drawing from the same distribution hoping for a different shape.
A useful cut-off: three attempts. If all three fail in the same way, the cause is in the input rather than the draw. Either the source photo cannot support the edit, or the instruction contains a judgement rather than a change.
That rule also protects a free allowance. Burning six generations on a photo that was never going to work is how people conclude a tool is bad when the photograph was.
The two habits underneath all five
Keep the original. It is your only diagnostic tool.
Change one thing at a time. It is the only way to know what any change did.
Adopt those two and four of the five mistakes become difficult to make. The full beginner sequence is in AI photo editing for beginners.
Four of Kitana's tools run in the browser, which makes the one-change-at-a-time habit cheap to practise; the thirteen photo tools are in the apps.
Frequently asked questions
- What is the single most common mistake?
- Asking for a judgement instead of a change. 'Make this look professional' is not a visual instruction — the model resolves it by moving toward an average it has seen labelled that way, which for a portrait is a younger, more symmetrical, less specific face than yours.
- Why does changing one thing at a time matter so much?
- Because when a combined instruction produces a bad result you cannot tell which part caused it. Separating the steps costs one extra generation and turns guesswork into diagnosis. It also stops a broad clause quietly altering something you did not intend.
- How do I know if a tool changed something it should not have?
- Compare specific features against the original at full size, not your overall impression — which is unreliable and biased toward whichever image you saw most recently. For a face: eye spacing, nose bridge, hairline, jaw, skin tone.
- Is starting from a bad photo really that limiting?
- Yes, and it is the mistake people most resist. These tools propagate and refine what is there; they do not recover what was never captured. Five minutes retaking a photo beats an hour of editing a poor one, every time.
- Why do results look worse after posting than on my phone?
- Platform compression. It targets exactly what heavy editing adds — fine high-contrast detail and saturated areas. Judging an edit at full screen and then publishing it somewhere it appears small and recompressed is the fifth mistake on this list.
- Are these mistakes specific to AI tools?
- Three of them predate AI entirely — starting from a weak source, over-editing, and judging at the wrong size are as old as photography. The two that are new come from generative tools failing confidently rather than visibly.
Ready to try it yourself?
Download Kitana and create your first AI photo in under a minute.