The best AI model for restoring an old photo depends on what is wrong with the image. A faded family portrait, a scratched print, a black and white photo, and a small low resolution scan are four different problems. One model rarely handles every problem equally well.

PhotoScanner therefore uses a selection of specialized image models from fal.ai. All four models covered in this guide are available directly in PhotoScanner for Mac. You can scan a photo, choose the right type of enhancement, review the result, and continue organizing your collection without moving files between several services.

Which AI Photo Restoration Model Should You Use

Choose All in One when a photo has several problems, such as fading, weak contrast, minor damage, and soft facial detail.
Choose Repair when scratches, dust, folds, tears, or other visible imperfections are the main problem.
Choose Colorize when you want to add plausible color to a black and white photograph.
Choose Upscale when the scan is too small or lacks enough resolution for viewing and printing.

These tools can also be used as stages in one restoration process. For example, you can repair physical damage first, colorize the cleaned image, and upscale the final version afterward.

Best Overall Model, fal.ai Photo Restoration

PhotoScanner calls this option All in One. It uses the fal.ai Photo Restoration model, which is designed to restore old or damaged photographs, remove imperfections, add color, and preserve the original character and details of the image.

This is the best starting point when a photo has several moderate problems at once. A faded portrait may need stronger contrast, cleaner facial detail, and small marks removed. Instead of treating every issue separately, the model produces a balanced restoration in one pass.

All in One is especially useful when you are processing a large family collection and need a reliable first result quickly. It can save considerable time, but the result should still be reviewed. If part of the original image is no longer visible, the model reconstructs a plausible detail. It does not recover information that has physically disappeared.

Best Model for Scratches and Damage, BRIA FIBO Edit Restore

PhotoScanner’s Repair option uses BRIA FIBO Edit Restore. According to fal.ai, the model automatically reduces noise, improves blur, enhances damaged photos, and removes imperfections while preserving the character of the original.

This makes it a strong choice for scratches, dust, fold marks, small tears, and worn areas. It is also useful when the photo already has acceptable color and you want to concentrate on visible damage instead of changing the entire look.

Repair models work by estimating what should appear beneath a damaged area. A thin scratch across a jacket can often be replaced convincingly because the surrounding fabric provides enough context. A large missing section of a face requires much more interpretation, so it deserves a closer review.

Best Model for Colorization, Topaz Adjust Colorize

PhotoScanner’s Colorize option uses the Colorize model in Topaz Adjust Image. The model is built specifically to add color to black and white photographs while retaining the source resolution.

Colorization can make people, clothing, landscapes, and interiors feel more immediate. It works particularly well when the original scan has clear tonal detail. A clean grayscale image gives the model useful information about light, shadow, and material.

The colors are plausible, but they are not historical evidence. An AI model cannot know the exact color of a dress, a car, or a wall unless that information is visible elsewhere. Keep the black and white original, and treat the colorized version as an interpretation of the scene.

Best Model for Resolution, SeedVR2

PhotoScanner uses SeedVR2 for its Upscale option. SeedVR2 can enlarge an image by a chosen factor or work toward a target resolution. PhotoScanner configures it for a high resolution result that is useful for larger displays and prints.

Upscaling is most valuable for small prints, old digital copies, and scans that do not contain enough pixels for the intended output. It increases the image dimensions and improves the appearance of edges and fine detail.

Upscaling should usually be the final AI step. Repairing scratches before enlargement prevents the damage from becoming larger and more prominent. It is also important to remember that an upscaler creates plausible detail. It cannot reveal a face or an object that was never recorded clearly in the source.

Can AI Extend a Cropped Photo

Repairing a missing corner inside the existing frame and extending the image beyond its original border are different tasks.

A restoration model can rebuild a damaged area within the scan. Extending the canvas requires a process called outpainting. The fal.ai Outpaint model can generate new content to the left, right, top, or bottom of an image. This can be useful when a photo was cropped too tightly or when a different aspect ratio is needed.

However, any newly extended area is generated content. It may look natural, but it was not captured in the historical photograph. Outpainting is also not part of PhotoScanner’s current AI toolset. PhotoScanner currently focuses on restoring the scanned frame through All in One restoration, Repair, Colorize, and Upscale.

A Practical Restoration Order

For a photo with several problems, the order of the operations can improve the final result.

Scan the original at the highest useful quality and correct the crop on the Mac.
Use Repair for obvious scratches, dust, tears, and folds.
Use All in One if the cleaned image still needs broader restoration.
Use Colorize if you want a color interpretation of a black and white photo.
Use Upscale last when you need more resolution for a display or print.
Compare the result with the original, especially around faces, hands, text, and important historical details.

Not every photo needs every step. A clean black and white portrait may only need Colorize. A faded color print may look best after a single All in One pass. Choosing the smallest useful intervention often produces the most authentic result.

Why PhotoScanner Combines Several Models

Many AI photo tools are built around one general enhancement. PhotoScanner takes a different approach by making several specialized models available within the same digitizing workflow.

You can capture prints with an iPhone or a connected flatbed scanner, correct the crop precisely on the Mac, apply the appropriate AI model, add dates and locations, and export the finished photos together. The original context of the collection stays connected to the restored images instead of being lost across separate websites and downloads.

This becomes especially valuable for larger collections. You do not need to decide which external service to use for every image, create several accounts, or repeatedly export and import files. You choose the problem you want to solve, then PhotoScanner sends the image to the appropriate model through fal.ai.

The Best Model Is the One That Matches the Damage

For mixed fading and damage, fal.ai Photo Restoration is the strongest general starting point. For scratches and physical defects, BRIA FIBO Edit Restore offers a more focused repair. Topaz Adjust Colorize is the right choice for black and white photos, and SeedVR2 is designed for higher resolution output.

Together, these models cover the restoration jobs most family collections actually need. PhotoScanner makes them available where they are most useful, directly inside the scanning, editing, organizing, and export workflow on Mac.

The Best AI Models for Restoring Old Photos in 2026