Pick for the damage, not for the star rating
Every list of the best photo enhancer apps ranks them one to ten, which quietly assumes they are all doing the same job in different amounts. They are not. A soft face, a small file, a motion blurred shot and a grey scan are four unrelated problems, and the model that answers one of them has nothing to say about the others.
So the ranking question is the wrong one to start with. Work out which of the four jobs your photograph actually needs, then judge an app on that job alone. This page is organised that way, including an honest note on where each class of tool fails, which is the part a listicle published by one of the apps tends to leave out.
Enhancer is four jobs wearing one word
Nearly every app in this category is a front end over one or more of these. Knowing which is which is what lets you read a feature list and predict whether it will help your picture.
- Face restoration. A model trained on faces rebuilds the features of a face that has gone soft or small. This is the one that produces the dramatic before and after shots, and it is the only one of the four that is specialised on a subject rather than on an image property.
- Upscaling. The file is enlarged and detail is proposed for the new pixels. Useful on anything small, whether that is an old scan, a crop or a picture saved at a low size years ago.
- Deblur and denoise. Softness and grain are reduced. Worth separating from upscaling in your head, because a large blurry file and a small sharp file want opposite treatments.
- Colourisation. Colour is proposed for an image recorded in grey. It is the most visibly impressive and the most obviously a guess, since the information was never captured in the first place.
Where each class fails, which is the part nobody lists
None of these are failures of a particular app. They are properties of what the underlying job is being asked to do, so they follow the capability from one product to the next.
- Face restoration on a face that is very small or very degraded produces a convincing person who is not quite the person. The model has learned what faces look like, so when the evidence runs out it supplies a plausible face rather than stopping. Watch the eyes, the teeth and the shape of the mouth, which is where a substituted identity shows first.
- Face restoration also tends to smooth. Skin texture, freckles and fine lines are exactly the detail that reads as degradation to a model, and a face can come back younger and cleaner than the person ever was.
- Upscaling invents texture on surfaces that had none recorded. Fabric, foliage, brickwork and hair get a pattern that looks right at a glance and does not survive being looked at closely. Text and patterned fabric are the fastest things to check.
- Deblurring works far better on softness than on motion. A picture that was slightly out of focus has a recoverable shape underneath it. A picture where the subject moved across the frame has had different parts of the scene recorded on top of each other, and no amount of processing separates them again.
- Colourisation is confident about the things it has seen a great deal of, such as skin, sky and grass, and it guesses at everything specific. A uniform, a car, a particular dress. If the colour matters to you, it is worth writing down what the colour actually was before you look at the result, because the result is persuasive.
- None of these classes touch physical damage. A tear, a crease, a scratch or a missing corner is damage to the paper rather than to the image, and it is a different kind of tool and often a different kind of decision.
What Clara runs, stated plainly
A list written by one of the apps in the category is only worth reading if it says where that app loses, so here is ours, in the same terms as everything above.
Clara enhances images through one third party inference provider, fal.ai, running two open source models: CodeFormer for face restoration and Real-ESRGAN for upscaling. Both are published models rather than a proprietary secret, which means every limit described above applies to Clara exactly as written and not in a softened version.
What Clara does is faces that have gone soft, images that are too small, flat or faded scans, and black and white pictures that want colour. What Clara does not do is repair the paper. Tears, scratches, creases, stains and missing corners sit outside the pipeline entirely, so a torn print comes back sharper and still torn. The services comparison on this site covers what to do with those instead, and some of that answer is to hire a person.
Judge any app with one photograph, in about a minute
Star ratings measure how people felt about an app, mostly on their own easy photographs. You want to know what it does to your hard one, and that test is quick.
- Use your worst photo, not a good one. Every tool in this category looks excellent on an image that barely needed help, which is why the marketing examples all look alike.
- Look at the result at full size rather than at the thumbnail. Enhancement is designed to be convincing at a glance, and almost every failure mode described above is invisible until you zoom.
- Check a face against another picture of the same person, if you have one. Not for sharpness, for identity. That is the failure that matters and the one a single image cannot reveal.
- Check anything with a pattern or with writing in it. Invented texture shows up there long before it shows up on skin.
- Run the same photo through two tools before concluding anything about either. Two results that disagree sharply tell you the evidence in the original is thin, which is useful information about the photograph rather than about the apps.
Four questions worth more than the ranking
Once an app has passed the test above on your own picture, the remaining differences are rarely about model quality at all.
- What size comes back? An enhancement returned smaller than your original has undone the part you cared about, and this is the most common quiet limitation in the category.
- What happens to the photograph after it is processed? Family pictures are not neutral data. It is a fair question to ask of any tool, and the answer should be in a privacy policy rather than in a review.
- Does the original survive? Anything that edits in place rather than writing a new file is asking you to trust a single pass with no way back.
- Can you tell what it did? A tool that applies four jobs at once on one button is impossible to debug when the face comes back wrong, because you cannot tell which stage caused it.
The input sets the ceiling on all of it
The largest single improvement available to most people is not a better app. It is a better capture of the original, because every tool here is working from what you hand it and none of them can recover what the scan never recorded.
A print photographed quickly on a phone at an angle, in a room light, carries a glare, a keystone and a colour cast that the enhancement will now faithfully sharpen. The same print captured flat, evenly lit and at the highest setting available gives every model on this page far more to work with, and the difference is usually larger than the difference between any two apps in the category.
If your issue is size rather than softness, the resolution guide covers what enlarging can and cannot recover. If it is blur, the unblur guide separates the two kinds and explains why one of them is answerable and the other is not.
More guides
Questions
What is the best photo enhancer app?
It depends on which of the four jobs your picture needs, since face restoration, upscaling, deblurring and colourisation are unrelated capabilities and no single ranking covers all four. Sort your photograph first, then judge candidates on that job with your own worst image.
Do AI photo enhancers change what someone looks like?
They can. A face restoration model supplies plausible detail where the evidence has run out, so a very small or very degraded face can come back as a convincing person who is subtly not the same person. Compare the result against another photograph of them, looking at the eyes and the mouth rather than at the sharpness.
Can a photo enhancer app fix a torn or scratched photo?
Clara does not, because that is damage to the paper rather than to the image. Some restoration software attempts it with results that depend heavily on where the damage falls, and damage running through a face is the case that goes worst. The services guide covers when this is a job for a person.
Are free photo enhancer apps good enough?
Often yes for the job itself, since several of the strongest models in this category are open source and widely deployed. The differences that matter tend to be elsewhere: the size of the file you get back, what happens to your photograph after processing, and whether you can tell which operation produced the result.
What is the difference between enhancing and restoring a photo?
Enhancing recovers information that is still in the file in a degraded form, such as softness, small size, fading or missing colour. Restoring puts back information that is physically gone, such as an area lost to a tear. The first is what an app does in a second and the second is a judgement call about what used to be there.