How to Colorize an Old Photo With AI and Get Good Quality
How to Colorize an Old Photo With AI and Get Good Quality
What the quality of old photo colorization depends on and how to prepare the picture
To colorize an old photo with AI and get good quality, you start with the source, not with the color. The print is digitized large and straight, the borders are cropped off, the damage is assessed, and only then does the file go to the neural network. The reason is simple: the AI colors what it sees. Light noise and fine scratches the tool will clean up by itself, but it may take a large crack for a real detail and paint it, turn a stain into a colored blotch, and lay muddy color over a blurry face. The cleaner and larger the black and white base, the more believable the color version will be.
How to colorize an old photo with AI: the short answer
Four steps: digitize, pick the tool that fits the condition of the print, get the color, check the result. The first step affects the outcome more than any other.
- Digitizing. A scanner or a careful phone shot, with no glare and no tilt, and with some size in reserve.
- Choosing the tool. An intact or lightly worn print goes straight to colorization. A print with cracks, tears and stains goes to restoration, which hands a black and white picture back already in color.
- Color. You upload the file to the Colorize photo tool: processing starts on its own, there is nothing to set up.
- Checking. You compare the faces with the original, look at the skin and the color edges, and upscale the file for printing if needed.
If the print is in good shape and was scanned large, it all comes down to uploading and checking. But old photographs like that are the minority: usually the print has at least some dust, a web of fine cracks or lost contrast, and how much of it there is decides which tool to choose.
Restore and colorize an old photo: do you need both tools
Usually not, one is enough. Colorization does more than add color: along the way it cleans the picture a little, meaning it tones down noise and removes fine scratches and slight blur. For a photo that has simply faded and got a bit rubbed, that is sufficient.
For badly damaged pictures there is a separate restoration tool. It repairs scratches, cracks and stains, redraws the faces, and returns a black and white picture already colorized. So there is usually no point in running the same photo through both tools: the second pass will use up one more free attempt and redraw the face one more time.
A simple question helps you choose: what bothers you most in the picture. If it is only the lack of color, take colorization. If it is a white scratch across a cheek, a crease through the whole frame or a water stain on a face, take restoration. When in doubt, compare two free previews made from the same scan.
What happens if you run the photo through both tools in a row
Most often the result is not better, just further from the original. Every pass through a neural network changes the drawing of the image a little, and this shows most on faces. After two passes in a row a person may look less like themselves than after one.
There is a practical side too. You would have to bring the clean file from the first step to the second, because the preview is reduced in size and covered by a watermark. So the rule is simple: pick one tool according to the condition of the print, and if you do not like the result, improve the original scan and repeat with it rather than processing what has already been processed.
What ruins colorization: a small scan
A small scan hurts colorization more than anything else, because it simply has no details for the neural network to hold on to. A face a few dozen pixels across is almost a blob to it: the eyes, the lips and the hairline merge, and the color goes on as one flat tone.
Typical sources of small files: a copy forwarded through a messenger, a screenshot, a photo of the print taken from a distance together with the table and the frame. If you have access to the original, rescan it larger: that will give you more than any later processing. If there is no original, it is worth upscaling the file with AI first and colorizing it afterwards.
What ruins colorization: scratches, creases and stains
Large scratches, creases and stains spoil colorization because the neural network may take them for part of the scene. It does not know that a white line came from a fold in the paper and was not in the picture from the start. Fine dust and thin scratches colorization usually smooths out by itself, with large defects you are not always that lucky.
- Deep scratches and cracks in the emulsion may stay as light lines, and on a color image they stand out more than on a gray one.
- Creases produce a band with a jump in brightness, and along it the color sometimes changes abruptly.
- Stains from moisture, glue or mold may be painted a random color and look like dirt on clothes or on a face.
- Torn corners and missing fragments colorization cannot fill in reliably: you will get a blank or a guess in their place.
If there are many defects like these, upload the picture to restoration instead of colorization: it is built for damage. How to deal with different kinds of damage is covered in detail in the article about scratches and creases on an old photo.
What ruins colorization: fading and weak contrast
A faded print comes out pale in color because it has little difference between light and dark. If the whole image is gray, with no true black and no true white, the color version will also be washed out, as if under a layer of dust.
Faces suffer most: on a faded print they almost blend into the background, and the neural network gives them a weak, hesitant tint. So before colorizing it is worth bringing the contrast back with a simple Levels or Curves adjustment in any editor. The important thing is not to overdo it: if the highlights are blown out to pure white and the shadows are crushed to black, color has nothing to hold on to there either.
Do you need to remove yellowing and sepia before colorizing
No, you usually do not need to desaturate a yellowed or sepia toned print on purpose. The tool converts the uploaded file to grayscale by itself before passing it to the neural network. This is deliberate: if a color cast remains in the input, the model takes it for real color, and colored bands and streaks appear in the result.
It does not depend on the file size: every uploaded picture is desaturated, small or very large. Saving the scan as gray in advance is not required, though it does no harm either. Your time is better spent on something else: cropping the borders and the frame out of the shot and making sure there is no glare on the scan.
An important thing follows from this. If you upload a faded color photo instead of a black and white one, its former colors will not be kept: the picture will turn gray and be colored from scratch, the way the neural network decides. For old color prints that have shifted to red or yellow this is a way to get a fresh version, but not a way to bring back the original colors.
Can you colorize an old black and white photo right away, with no preparation
You can, and for a first rough look that is exactly what you should do. Processing and the preview are free, no registration is needed: upload the file as it is and see what comes out. If the color went on cleanly, no preparation is needed, and if you see colored blotches and dirt on the faces, upload the same scan to restoration and compare.
Keep two things in mind. The preview is reduced to 900 pixels on the long side and covered by a watermark, so it shows the overall color and the big mistakes, but not the fine details. And there are not many free runs: five per day from one address, with one shared count for colorization, restoration and upscaling. Spend your attempts with some thought.
Step by step: digitizing, choosing the tool, upscaling, color
Below is each of the four steps on its own: what to choose in the form, what to look at and which file to pass on. Not every picture needs the second and third steps.
Step 1. How to digitize an old photo for colorization
The best option for colorization: a flatbed scanner, the print lying flat on the glass, the resolution chosen with room to spare. A scanner gives even light and does not distort the geometry, and that is exactly what a neural network needs.
Scan large, especially small prints. A print the size of your palm scanned at low resolution turns into a file where the face takes up a couple of hundred pixels. Save without heavy compression: JPG at maximum quality or PNG. Color or grayscale mode makes no difference for colorization (the tool desaturates the file by itself), but a color scan is more useful for the archive: it shows the real condition of the paper.
If you have no scanner, a phone will do. Put the print on a flat surface near a window, hold the phone strictly parallel to it, and make sure there is no glare and no shadow of yours on the glossy surface. A detailed guide is in the article on how to digitize old paper photos at home.
What to do with the scan before uploading: crop and rotate
Before uploading, the scan needs to be cropped to the edge of the image and rotated straight. The neural network processes the whole frame, and everything extra takes attention away from the main subject: white borders, an embossed mount, the edge of the table, neighboring prints on the album page.
Borders and background also harm the color. A cardboard backing may get a random tint, and from it the color often creeps onto the edge of the picture itself. If there are several photographs on an album page, cut each one out into a separate file and colorize them one at a time.
Check the format as well. The tool accepts JPG, PNG and WebP. If your phone saves pictures in another format, for example HEIC, convert the file to JPG first.
Step 2. Restore or colorize: which tool to choose
If the picture has scratches, cracks, stains or blurred faces, use the separate Restore photo tool instead of colorization. The neural network removes the damage, finds the faces and redraws them, and returns a black and white or toned picture already colorized. There is usually no need to color it separately after that. It accepts the same files as colorization: JPG, PNG, WebP up to 25 MB and up to 8000 pixels per side, and it delivers the result as a JPG.
There is no processing mode to choose in the form, the only thing to choose is upscaling: “No upscaling”, ×2 or ×4. The default is ×2, so the restored file comes out twice as large as the source right away, although for large scans the final size is capped by the server. For a small scan this is convenient: restoration and upscaling happen in one go. If the scan is already large, switch to “No upscaling”.
An honest word about the limits. The neural network draws the restoration result anew, so it is a reconstruction, not a copy of the print. Where the image is lost (a torn corner, a scratch across the whole face), it invents what it cannot see, and on a heavily blurred portrait it may change the features. Compare the result with the original and make sure the person still looks like themselves. If the likeness is gone, that is a job for manual retouching. And an intact picture with no visible damage does not need restoration: the colorization from step 4 is enough for it.
Step 3. When to upscale a photo before colorizing
Upscaling is worth it if the faces on the scan are small or if you plan to print larger than the original print. On a large file the neural network makes out more details: the edge of the lips, the iris, the folds of fabric, and the color goes on more neatly.
There is a separate tool for this, Upscale photo: scale ×2 or ×4, “Universal” and “Fast” modes, result in JPG. If you already enlarged the picture during restoration, there is no need to upscale it a second time. And do not upscale everything to the maximum. Upscaling takes in no more than 2000 pixels on the long side: a larger picture is first reduced to that size, so the result does not exceed 4000 pixels at ×2 and 8000 at ×4. Colorization, in turn, works with a frame no larger than 3000 pixels on the long side: it will accept a bigger file (up to 8000 pixels per side and 25 MB), but will reduce it before processing. Pushing a scan to 8000 pixels for the sake of color is pointless.
A practical rule: if the long side of the scan is under one and a half to two thousand pixels, upscaling 2 times is usually justified. If the file is already large, you can skip the step: colorization does not use more than 3000 pixels on the long side anyway.
Step 4. How to colorize an old photo online with AI
Colorization takes the fewest actions: just upload the file and processing starts on its own. Colorization has no settings. One file at a time, by dragging it into the window or through the file picker, with no registration.
- Upload the prepared black and white file.
- There is nothing to choose or press: processing starts on its own as soon as you select the photo or drag it in.
- Wait for the result. Processing runs on the server, and the time depends on the file size and the load.
- Judge the preview: skin color, clothes, background and how closely the faces match the original. Not happy: improve the original scan and upload it again.
The result is saved as a JPG, no larger than 3000 pixels on the long side. The preview is free, it is reduced to 900 pixels on the long side and carries a watermark. The clean full size file is paid. There are no sliders for adjusting color in the tool: it raises saturation slightly by itself, cleans up a bit of noise and fine scratches along the way, and fine tuning is done in the finished file.
Colorize a black and white photo
AI adds believable colors: skin tones, sky, greenery, clothing and city scenes. Local, no subscriptions, no watermark.
A family archive? Manual colorization
For historically important photos, AI colors are a starting point. Manual colorization layer by layer, taking the era and the tones of skin, fabrics and background into account, gives a result that fits print and archive better. From $10 per shot.
Order manual colorizationWhich file to pass from step to step
To each next step you need to bring the clean file from the previous step, not the preview. The preview is reduced and covered by a watermark: the mark will become part of the image, the neural network will colorize it along with everything else, and the small size will cancel out all your preparation.
A sensible sequence follows from this. First run the original scan through colorization and through restoration separately, for free, and look at the previews. That way you will see which of the two tools handled your particular picture better. Only once you have decided do you get the clean file and, if you need a large print, send it on for upscaling.
| Step | What you get | What to watch for |
|---|---|---|
| Digitizing | A scan or a photo of the print | JPG, PNG or WebP, no borders, no glare |
| Restoration | A cleaned file, a black and white picture arrives already in color | Faces look like the original, no “plastic” skin |
| Upscaling | The same file 2 or 4 times larger | 3000 pixels on the long side is enough for colorization, size up to 25 MB |
| Colorization | A color JPG, lightly cleaned | Likeness of faces, skin, color edges, areas left uncolored |
Enhance and colorize an old photo online: can it be done in one action
To a large extent, yes. Colorization improves the picture a little along the way: it tones down noise and removes fine scratches and slight blur. Restoration repairs serious damage and hands a black and white picture back in color right away. For most prints one of these two tools is enough. The only thing left as a separate action is upscaling: it works with the drawing of the image and adds no color.
The convenience has a downside. When cleaning and color are done with a single click, you do not see the intermediate result and cannot keep one without the other. Along with the noise, the neural network may slightly change small facial features: the shape of the eyes, the line of the lips, the form of the nose. So the finished picture must always be compared with the original, portraits above all.
In practice the path looks like this. For an intact print with a good scan, colorization alone is enough. For a worn, scratched or torn print, people take restoration and get the color at once. Upscaling is added for small scans and for pictures that are going to be printed large.
How many pictures can you process at once
Colorization works one file at a time, there is no batch upload, and the next run can only be started after the previous one has finished. Five runs per day are available for free from one address, with one shared count for colorization, restoration and upscaling. That is enough to compare colorization and restoration on one picture and to repeat a couple more times with an improved scan. It is not worth spending attempts on uploading the same file again: colorization has no settings, and the result will not change. After that the site will offer payment, and the free attempts will not cover an album of dozens of prints.
If the service replies that it is busy, either your previous run has not finished yet or the server is loaded with heavy tasks at that moment. Wait a couple of minutes and try again.
Do not want to fiddle with it? Hand the photo to a retoucher
Automation does not fix everything: a difficult frame, a photo that matters, an old print off a scanner. Where the algorithm gives up, a person works the picture layer by layer in Photoshop, by hand.
BeforeAfter⇆- Bring back sharpness in the face and eyes
- Remove noise and artefacts without plasticising the skin
- Fix colour, contrast and a broken exposure
- Remove scratches, stains and creases from an old scan
- Return a file large enough to print
- Price per photo$10
- Turnaroundwithin 24 hours
- Deliveryto your email
- If we cannot do itmoney back
We will finish your photo by hand
A studio retoucher fixes what automation cannot: sharpness, face, colour, defects. Pay first, then simply send the photo and tell us what you need. If we cannot do it, you get your money back.
- Manual photo retouching$10
How to judge the quality of colorization and what to do with the result
A colorized picture should be checked before you pay for the clean file or take it to print. The check takes a couple of minutes and follows three lines: whether the faces look like the original, whether the color sits evenly on the skin, and whether it has crossed the outlines. Colorization has no settings at all, so one main lever remains: the quality of the original scan.
Colorizing an old photo in good quality: what it depends on
The quality of colorization depends on two things: the source file and how guessable the content of the picture is in the first place. You can influence the first, but not the second.
The source file matters most. A large, sharp, contrasty scan with no defects gives a good result on the first try. A small and dirty one cannot be saved by any amount of reprocessing. So when you want “good quality”, put the effort into preparation: a larger scan, even light, no glare and no borders. And remember that real quality is visible only in the clean file: the preview is reduced to 900 pixels.
The content sets the ceiling: a portrait against a garden the neural network will color convincingly, while a shop window full of goods or a busy patterned carpet will most likely come out approximate.
Colorization has no settings: what to do if the color looks wrong
If you are not happy with the color, the thing to change is the source. Colorization has no switches and no parameters: you select the photo or drag it in, and processing starts right away on its own. So uploading the same file again will give you almost nothing: the image will most likely come out exactly the same.
Each run is still deducted from the five free ones per day, which are shared by all the heavy tools on the site. Save your attempts for an improved scan or for a comparison with restoration. What really changes the result: a larger scan, even light, no glare and no borders. And if the important colors are still wrong, what remains is manual finishing by a retoucher.
How to compare a colorized face with the original
Open the original and the result side by side and look at the features, not at the color. Colorization also cleans up noise and blur, and on a small or soft face the shape of the eyes, the line of the lips, the form of the nose and the apparent age may shift slightly along with them.
Check in order: eyes and eyebrows, nose, mouth, the oval of the face, the hairline. Moles, scars, wrinkles and glasses should stay where they were. A good method: show both pictures to someone who knew the person and ask whether they recognize them in the color one. For a memorial portrait or a picture that will be used for a document or a plaque, likeness matters more than color. If it is gone, you need the manual work of a retoucher.
Wrong color of clothes or a uniform: what to do
Running it again will not help here: the neural network does not know what color the dress was, and on the same file it will give the same answer. The fix has to be made in the finished picture. A single mistake, for example a blue dress instead of a burgundy one, can be corrected in an editor with layers: select the object and change the hue in that area only.
Uniforms, medals and banners are harder: they have many small details with a strictly defined color, and those have to be checked against reference materials. It makes more sense to order such a correction from a specialist and tell them at once everything you know about the picture: the branch of service, the years, the person's eye and hair color.
Colorizing an old portrait: what to look at first
In a portrait you look at the face first: at the likeness, then at the skin color, and only after that at everything else. A viewer will easily forgive the clothes and the background, but someone else's features or a face of a strange tint catch the eye at once.
- Close-up portrait: compare the eyes, the lips and the oval of the face with the original.
- Group photo: check that all the faces are colored equally evenly and that the small ones in the back row have not changed beyond recognition.
- Landscape, architecture, street: look at the sky, the greenery and long surfaces of a single color.
- A picture in military uniform or in national costume: the neural network does not guarantee correct colors, more on that below.
How to remove an overall color cast in any photo editor
An overall color cast is removed after colorization, in the finished file. The tool itself has no settings for this, and uploading the same picture again will not change the tint, so there is no reason to spend a free attempt on it.
If the whole image looks as if it were behind colored glass, warm or cold, it is more reliable to fix that in the finished file. You need two sliders that every editor has, including the one in your phone: temperature (warmer or colder) and tint (greener or pinker). The reference point for the adjustment: whatever in the picture is known to be white or gray, such as a shirt collar, gray hair, paper or a whitewashed wall, should look neutral.
How to judge the quality of colorization: skin and faces
The quality of colorization is judged by the skin first of all: it should be one even tone for each person, with no gray or greenish islands. Open the result and go down the list.
- The face, neck and hands of one person match in tone.
- There are no gray areas left without color on the forehead, under the eyes or at the hairline.
- The lips differ from the skin but do not look painted, unless they were in the original.
- The whites of the eyes and the teeth have stayed light and have not turned pink along with the skin.
- The hair has not merged with the background in color, and gray hair is still gray.
If the skin came out blotchy, the cause is almost always in the source: noise, stains or blur on the face. That is a signal to improve the scan or to try the restoration tool, not to run the same file again.
How to judge the quality of colorization: color edges
The second sign of good colorization: the color does not cross the outlines. Look at the places where different objects meet: a collar and a neck, a hand against a dress, a head against a wall, a child in an adult's arms.
A bad sign: skin color bleeds onto the collar, the color of the dress tints the hand, the green of the background settles on the hair. This happens at blurred edges and on small details. On a large, sharp file there is noticeably less bleeding, so upscaling before colorizing really does help here.
Check long surfaces of a single color separately: a wall, the sky, a coat. The color on them should be continuous. If half of the coat is blue and half is gray, the neural network did not understand that it is one object. An area like that is corrected in an editor or handed to a specialist.
How to check a colorized photo before printing
Before printing, look at the clean colorized file at full size, not at the preview and not at a thumbnail on a phone screen. Reduced in size, any colorization looks decent, and the flaws are visible only when you zoom in to actual pixels.
Go over the faces, the hands, the edges of the clothes. Then step back and judge the overall impression: does it feel like a hand tinted postcard or, on the contrary, like a dull picture. Compare it with the black and white source: the color version should not lose detail in the shadows and highlights, and the faces should stay recognizable.
And keep the paper in mind: on a print, colors usually look a little darker and denser than on a glowing screen. Slightly restrained colorization looks more natural in print than a very saturated one.
What to do if the colorization came out badly
If you are not happy with the colorization, change the source, since colorization has no settings, and one thing at a time. That way you will see what exactly affects the result and will not waste your free attempts.
- Crop off the borders, background and frame if they are still in the shot.
- Rescan the picture larger, in even light and with no glare.
- Pale result: raise the contrast of the source.
- Stains, scratches and dirt on faces: upload the scan to restoration, it will return the picture in color right away.
- Small faces and color bleeding: upscale the file and colorize it again.
- Overall color cast: adjust temperature and tint in the finished file.
- The face no longer looks like the person or important colors are wrong: manual finishing is needed.
There is little point in uploading the same file again: the result will most likely not change. And if after all the fixes the important colors are still wrong, it is no longer about the quality of the file but about the fact that the neural network cannot know the truth about the picture. The next section is about that.
What AI guesses right and what it cannot know
A neural network does not recover the real colors of an old picture, it picks plausible ones. A black and white file holds no color information at all: two dresses, a red one and a green one, may look equally gray in it. The AI paints things the way they usually are in similar images, and in some cases that is a hit, in others a guess.
What AI guesses well when colorizing an old photo
Everything whose color in real life hardly changes from picture to picture is guessed well. Here the neural network has solid ground.
- Skin. Faces and hands, if they are large and sharp enough.
- Sky and water. Blue and gray-blue shades.
- Vegetation. Grass, foliage, trees.
- Soil, sand, wood, brick. Natural and building materials with a recognizable texture.
- The overall light of the scene. A sunny day and an overcast one, indoors and outdoors are usually told apart correctly.
That is why street scenes, landscapes and outdoor portraits come out most convincingly: most of the frame is taken up by things that are hard to get wrong.
What the neural network cannot know: the color of clothes, eyes and hair
The color of clothes, eyes, hair, wallpaper, a car or a toy is something the neural network cannot know. It picks the most likely option, and that often turns out averaged: clothes drift toward gray-blue and brown tones, eyes toward dark ones.
If you remember that your grandmother had a burgundy dress and blue eyes, be ready for the colorized picture to show something else. This is not a defect and not poor file quality: the right answer is simply not recorded in the image. No amount of scan preparation will help here.
Hair is a little easier: dark and light differ in brightness, and so does gray hair. But red cannot be told from light brown in a black and white picture, and the neural network will choose something neutral.
Military uniforms, medals, flags: why AI gets them wrong
Uniforms, insignia, medals, flags and signs are colored by the neural network at random, and the mistakes here hurt the most. These objects have a strictly defined color, yet in the picture they take up little space and consist of small details.
A medal ribbon may come out in a single color, shoulder boards may merge with the tunic, a banner may end up the wrong color. For looking at with the family this is often acceptable, but for a display, a museum, a publication or a memorial portrait of a veteran such colorization has to be checked against reference materials and corrected by hand.
Group photos and small faces: what to expect
In a group picture the quality of colorization drops with every row: the large faces in front come out well, the small ones further back come out worse. A small face may stay grayish or pick up the tint of a neighboring object.
Upscaling the file before colorizing helps. Another trick: cut the person who matters out of the group as a separate frame and colorize them as a portrait. The face will take up most of the image, and the result will be noticeably better than in the full picture.
Does AI change the face when colorizing
It can, though usually only a little. Colorization does more than add color, it also lightly cleans the picture of noise and blur, and on a face this sometimes touches the small features. The larger and sharper the face on the scan, the less it is affected. Small faces in group pictures and blurred portraits are at the greatest risk.
Restoration interferes even more noticeably: there the neural network redraws the faces. So likeness is checked after any processing, by comparing the result with the original. For important portraits, meaning a picture for a document or in memory of a person, this check is a must. If the person in the color version does not look like themselves, running it again will not fix that: manual work is needed.
Colorize an old photo online free and without registration: what is free and what is not
Processing and the preview are free, and no registration is needed for them: you upload a picture and get the colorized result, reduced to 900 pixels on the long side and with a watermark. What is paid is the clean file at full size and without the mark. There are five free runs per day from one address for all the AI tools on the site together, so you can compare colorization and restoration on one picture.
This is convenient precisely for old photographs, where the outcome is not known in advance. You see how the neural network handled your picture and pay only when you are happy with the result. If you are not happy with it, you lose nothing.
One note about privacy, so there are no illusions: processing runs on a server, the photograph is sent there for that purpose, and the finished result is kept there for some time so that it can be downloaded after payment. If the picture must not be passed on anywhere, an online tool is not your option.
Which files the tool will not accept
The tool will not accept a file of the wrong format, one that is too heavy, or one with an unsuitable size in pixels. The exact limits are these:
- formats: JPG, PNG and WebP only;
- file size up to 25 MB;
- dimensions no smaller than 100 by 100 pixels;
- each side no larger than 8000 pixels.
Convert iPhone pictures in HEIC format to JPG first. The same limits apply in the restoration and upscaling tools.
The lower limit is a formality: a file a hundred pixels across will upload, but no good colorization will come out of it. Aim for the face of the main person to be at least a few hundred pixels high.
When you need manual colorization instead of AI
Manual colorization is needed when accuracy of color matters and not just a pleasant picture. A neural network gives a fast and plausible result, a colorist gives a correct one.
- You know the real colors and want them respected: a dress, the eyes, the color of a house.
- The picture shows uniforms, medals, banners, coats of arms.
- The photo is being prepared for a large print, a memorial, a book, an exhibition.
- The print is badly damaged, and AI restoration could not cope.
- After processing the person no longer looks like themselves.
On the tool page you can order manual colorization from a specialist. And if your case is simpler and the picture is in decent condition, a general overview of methods and apps is in the article on how to colorize a black and white photo online.
Can you fix the colors yourself after the AI
Yes, the finished color JPG can be refined in any image editor, and for a couple of small mistakes that makes more sense than ordering a full manual job. The tool itself has no color adjustment, so the editing is done in the downloaded file. Overall saturation, warmth and brightness are changed with ordinary color correction, which even a phone editor can handle.
Individual objects are harder. To repaint a dress from blue to burgundy, you need to select it and change the hue in that area only, without touching the hands and the background. This is done in editors with layers and masks and takes care at the edges. If many such fixes pile up, it is easier to hand the picture to a specialist right away.
What to keep after colorizing
After colorizing, keep the original scan and the color version, and if you tried restoration, its result too. A colorized picture does not replace the original, it is a modern reading of it.
You will come back to the original scan: to check the faces against it, to hand the picture to a specialist, or to colorize it again when a better tool appears. For that it has to be kept large and without heavy compression. Do not throw away the paper print either: it remains the only true source.
Frequently asked questions
Can I colorize an old photo with AI for free? Yes, processing and the preview are free and need no registration: you see the colorized result, reduced in size and with a watermark. There are five free runs per day from one address. Only the clean full size file without a watermark is paid.
Should I restore or colorize an old photo first? Usually you need one tool out of the two. Upload an intact picture straight to colorization: it also lightly cleans up noise and fine scratches. Take a badly damaged one to Restore photo: it repairs scratches and tears and returns a black and white picture in color right away. In both cases compare the faces with the original.
Do I need to convert a yellowed photo to black and white myself before colorizing? No. The tool converts every uploaded file to grayscale before processing, so that yellowing or sepia does not cause colored streaks. For the same reason the former colors of a faded color photo are not kept: it is colored from scratch.
Which settings should I choose when colorizing an old portrait? None: colorization has no settings. You select the photo or drag it in, and processing starts on its own. Uploading the same file again will not change the result, and the attempt will be deducted from the five free ones per day. If you are not happy with the color, it is more useful to improve the scan or order manual finishing.
Will AI show the real colors of clothes and eyes? No. A black and white picture holds no color information, so the neural network picks a plausible option. Skin, sky and greenery are usually guessed well, while the color of a dress, eyes, a uniform or medals may not match the real one.
Why are there gray or colored blotches on the face after colorization? Most often the source is to blame: noise, stains, scratches or blur on the face. Rescan the picture larger and without glare, and if the damage is noticeable, upload it to Restore photo: it repairs the defects and returns a black and white picture already in color.
What files can I upload for colorization? JPG, PNG or WebP up to 25 MB, from 100 by 100 pixels and no more than 8000 pixels on each side. The result is saved as a JPG. One file is processed at a time.
Can I colorize a photo with a torn corner or a large stain? A picture like that is better uploaded to Restore photo: it repairs scratches, stains and tears and delivers color right away. But the neural network invents a lost fragment, it does not bring it back. If there was a face or an important detail in that spot, manual retouching is needed.
We will edit a sample from your catalog for free
For online stores, manufacturers, brands and wholesale suppliers. Send 1-3 product photos and we will return them professionally edited, with a price quote for your whole catalog.
This is a b2b, volume format. We do not process personal photos (portraits, events, documents) here.
Request received
We will edit your sample and send the result with a catalog quote within one business day.