Complete Tutorial on AI Repair of Old Photos, 2026 Practical Guide to Mobile Phone and Computer Tools

Complete tutorial on AI restoration of old photos, 2026 Practical Guide to Mobile Phone and Computer Tools

Those old photos at home that are yellowed, blurry, scratched, and faded. In the past, if you wanted to repair them, you had to spend money to hire a professional photographer for post-processing. Now you can use AI tools to do it yourself. Whether it's a black-and-white photo of your parents when they were young, a faded color photo of your grandma, or a blurry photo taken a few years ago when your phone didn't have enough pixels, it can be restored to a clear and viewable version within a few minutes. This article explains clearly the commonly used AI solutions for repairing old photos in 2026, from mobile phones to computers, and from free to paid. It is suitable for readers with no basic knowledge to get started directly after reading this article.

What can AI do to repair old photos?

AI’s ability to repair old photos is better than many people think. What it can do is: improve the clarity of blurry photos so that the outline of the face can be recognised, automatically colorize black and white photos to restore the approximate color, repair scratches, mildew spots, creases and other defects, improve the resolution of old photos so that they are not blurry on the big screen, and restore severely faded color photos to a tone close to the original.

You should also know what you cannot do. AI cannot see the original data and can only make reasonable guesses based on pixel distribution. Therefore, the reconstructed face shape, clothing patterns, and background details are all imagined by AI and may not be 100% true to reality. If you expect a photo to be exactly what it looked like when it was originally taken, that's not possible. But if you want to print out the photos and hang them on the wall, make a photo album, or pass them on to your family as a souvenir, the AI ​​repair effect is completely sufficient.

A typical restoration process for an old photo

Restoring an old photo usually involves four steps. The first step is to scan or copy the paper photo to digitize it, use the mobile phone camera to align it vertically, make the light even to avoid reflection, and import it into the album. The second step is the initial repair of scratches and blemishes, which is the strength of AI tools. The third step is to improve the clarity, making the details of key parts such as faces and text clear. The fourth step is coloring or color restoration, turning black and white photos into color or restoring faded color photos to natural tones.

You can choose a different combination of tools for each step. Some tools are all-inclusive, while others are only good at one or two steps. The following introduces the mainstream solutions for mobile phones and computers respectively.

The first choice for mobile phones: Remini and domestic applications

Remini is the world's most famous old photo restoration app, and its core algorithm does a very good job at restoring facial details. An old photo that is so blurry that only the outline can be seen can be restored to a relatively clear level after Remini processing. It supports iOS and Android. The free version can edit several photos per day. Pro subscription is available on a monthly and annual basis. The specific price is shown in the app.

There are many domestic alternatives. Meitu Xiuxiu's old photo repair function, Qingyan Camera, and Awakening Picture all have built-in AI repair capabilities. The operation process is similar to importing photos, selecting repair mode, and waiting for a few seconds to output the picture. The advantages are the Chinese interface, no need to hang a ladder, and the price is cheap. The disadvantage is that the detail accuracy is slightly weaker than Remini. Not much of a difference for most everyday repairs.

Baidu's Repair Master applet and Alibaba's ModelScope repair model also provide online versions, which are completely free and suitable for people who don't want to download apps. Just search it on WeChat to use it.

PC: GFPGAN plus Topaz Photo AI

If you have more photos to batch process, or are pursuing more detailed repair effects, the tools on the computer will be more professional than those on the mobile phone.

GFPGAN is an open source face restoration algorithm from Tencent ARC Laboratory and is available for free. Technical parties can download it locally and run batches in Python. Ordinary users can use the web version of the tool based on the GFPGAN package, such as the ARC official demo page. Its restoration effect on facial details is recognized in the industry as very strong, and it is suitable for restoring close-ups of characters.

Topaz Photo AI is a commercial software, positioned in high-end photography post-production, focusing on resolution improvement, noise reduction, and sharpening. When repairing old photos, you can use it to maximize the clarity first, and then use other tools to color them. The advantage of Topaz is stable quality and many controllable parameters. The disadvantage is that it is not cheap and requires a Windows or macOS computer.

PS plus Generative Fill is also an idea. Adobe has integrated AI generated fill in PS, which can intelligently repair damaged areas and is suitable for local refinement. If you are already a PS user, this workflow has the lowest cost to get started.

Online free solutions: Alibaba Damo ModelScope and Tsinghua Hugging Face

If you don’t want to download any software, pure web page repair is the most trouble-free solution.

The ModelScope platform of Alibaba DAMO Academy has opened up multiple image restoration models, including GPEN face restoration, ESRGAN super-resolution, DeOldify black and white coloring, etc., which can be run directly by uploading images on the web page. The free quota is more than enough for individual users, and the Chinese interface is also friendly.

The Spaces area on Hugging Face has a large number of restoration model demos contributed by the community. You can find usable demos by searching for keywords such as photo restoration, colorize, super resolution, etc. The disadvantages are that the page is in English, there are sometimes queues, and the speed of drawing pictures depends on the server load. But it is very suitable as a free trial tool. You can compare the effects of several models on Hugging Face before deciding whether to spend money on Remini.

Tencent ARC Lab also has a public repair demo web page with a Chinese interface, which can be uploaded directly and is suitable for the elderly.

Black and white photo coloring tutorial

Colorizing black and white photos is the most dramatic step in restoring old photos. The effect of AI coloring is surprising, and it can often make a photo taken decades ago instantly warm.

The mainstream free tool is DeOldify, which is an open source algorithm and has demos on ModelScope and Hugging Face. It will intelligently predict the color based on the objects and scenes in the picture. Most of the clothes will be blue-gray, the skin will be natural flesh color, the sky will be light blue, and the plants will be green. The effect is generally natural, but the guess may be wrong for some uncommon objects, such as the specific color of military uniforms.

Online tools such as Algorithmia, Palette.fm, and Hotpot.ai all provide black and white coloring services. The free quota is limited, and the payment is per picture or monthly. If you want to repair dozens of photos, it is cost-effective to buy a one-month membership.

After coloring, it is recommended to manually fine-tune it. Adjust the skin color saturation and overall color temperature to a natural and comfortable range in the picture editor on your mobile phone or computer. This step can take the effect to a higher level.

Practical Combat of Clarity Improvement and Noise Removal

The core tools for improving clarity are Real-ESRGAN and SwinIR. The former is good at real details in photos, and the latter is good at cartoons and animation. These algorithms are integrated into packaged desktop tools such as Topaz Photo AI and Upscayl, and ordinary users can just drag and drop photos.

Upscayl is a free and open source desktop version, available for Windows, macOS, and Linux. It supports enlarging photos from 2 to 4 times. The processing speed depends on your graphics card. Particularly friendly to independent graphics card users, one image takes just a few seconds. The integrated display can also run, just a little slower.

In terms of noise removal, Topaz DeNoise and DxO PureRAW are commercial solutions with the highest accuracy. The free plans include Noise Reducer web version, Photopea plus plug-in, AI Image Enhancer and other tools, which are enough for daily use.

The practical suggestion is to remove noise first and then amplify. If the order is reversed, the noise will also be amplified into color spots. If you finish the two steps before painting or repairing, the process will be smoother.

After repairing, use AI drawing tools to complete the background or recompose the image.

In some old photos, the characters are clearly edited, but the background is blurry or the composition is skewed. At this time, you can use AI drawing tools to complete or recompose the image. Cut out the character part, let the AI ​​drawing tool generate a new background and then synthesize it.

Lingtu on iOS is an AI drawing available in China. The App integrates a Midjourney-style atmosphere engine, a Flux-style realistic engine, and a Nano Banana-style fast engine. You can directly search for spirit maps and download them without having to hang up a VPN. Match the restored old photo with a background image of the same era, or re-create a digital frame based on the tone of the original photo, and the presentation effect of the entire photo will be improved to a higher level. The Chinese interaction is very smooth. Even if you use it for your parents, you don’t need to learn English terminology. It’s worth a try.

On the computer side, you can use Stable Diffusion plus ControlNet, which has higher accuracy but a higher threshold, and is suitable for technical people. The lightweight solution on the mobile phone is sufficient for most home scenarios.

Fixed archiving and printing suggestions

Don't just keep the restored old photos in your phone. It is recommended to back them up to a cloud disk and an external hard drive at the same time. Save one copy each in iCloud, Alibaba Cloud Disk, and Baidu Cloud Disk, plus a mobile hard drive to store the original scan and repaired version. This set of three backups is a standard practice for family photo libraries.

If you plan to print out the photos and hang them on the wall as a photo album, choose a high-definition version with a resolution of 300dpi or above. You can place orders for hardcover photo albums or frame mounting on JD.com, Taobao, and local photo studios. A restored photo of grandparents when they were young is paired with a walnut photo frame and displayed in the living room. It is a very warm family gift.

The repaired electronic version can be made into a family digital photo album, placed in a NAS or cloud disk, and set as a link that the whole family can access, and elders can also browse it easily.

FAQ

Will the face restored by AI turn into a different person?

There will be certain differences, especially when the original photo is extremely blurry. The facial details reconstructed by AI are inferred based on the training data and may not be completely identical to the person himself. The general approach is to ask elders or family members who are familiar with the person to confirm whether the overall look is correct after repair. If the difference is too big, run it again with a lighter repair mode to retain the original outline features.

Is there any fee to restore old photos? Remini

If you only edit a few pictures, the daily quota of the free version is enough. If you have dozens of photos that need to be repaired in batches, you can buy a one-month membership to process them all together and then unsubscribe. The cost of dozens of dollars can solve a lifetime of old photo problems. If you don’t want to pay, the domestic Meitu Xiuxiu and Alibaba ModelScope can also achieve similar results, but the single-shot speed is slower.

Are the colors of black and white photos accurate after coloring?

Not necessarily accurate. AI coloring is guessed based on training data. The probability of correctly guessing common objects such as skin, plants, sky, and wood is very high, but the specific color of clothes and decorations depends on luck. After repair, it is best to let the elderly at home take a look. If you remember what color the dress was back then, you can manually adjust it to the correct hue in the editor.

What’s the best way to scan old photos?

If you have a flatbed scanner, which will produce the best scan quality, a resolution of at least 600dpi is recommended. If you don’t have a scanner, use your mobile phone’s camera to take the photo vertically, next to a window with even indoor lighting or under a white desk lamp, to avoid reflections and shadows. Apps such as Scanner and White Draw on your mobile phone have built-in perspective correction functions, which can automatically correct the angle to a head-on view after taking a photo.

Can restored photos be used commercially?

If the photos are your own private photos, they can only be used for family archiving, photo album printing, and social sharing after restoration. There is no problem at all. If you want to take restoration services as a side job, make sure the client has ownership and portrait rights of the original photos to avoid portrait disputes involving third parties. If the photo comes from the Internet or historical data, the copyright status of the original material must be confirmed before commercial use.