Important points
- Microsoft's free AI effectively recovers old photos with one click and outperforms Photoshop results.
- Restore, colorize, and upscale your old photos using Replicate's catalog of AI tools.
- Replicate's additional AI models allow you to further enhance and refine your photos, offering endless restoration possibilities.
Photoshop has built-in neural filters to restore old, scratched photos. But you don't have to spend $20 a month just to use that feature. Microsoft has developed a more robust AI for this job. This is completely free. This is the only guide you need to bring your old photos back to life.
Microsoft's AI brings old photos back to life
There was a time when Photoshop gurus needed hours of painstaking work to restore old photos that had deteriorated. Nowadays, AI can do most of the heavy lifting. You can even recover photos with just one click.
Bringing Old Photos Back to Life is Microsoft's official project for AI-powered photo recovery. A sample of the project shows an impressive restoration of several deteriorated photographs. I also tested it with my own photos and it worked beautifully.
Unfortunately, Microsoft has not yet provided an official app to use this AI. If you own a powerful PC and have the technical know-how, you can run AI on your own hardware.
But the easiest way is to run it for free on the cloud. You can use either:
Colab lets you run your code in the cloud using Google's hardware resources. We first tested Bringing Old Photos Back to Life AI on a Google Colab notebook. For me there were a lot of bugs. Mileage may also vary. It's a little difficult to use.
Replicate makes it easy to run open source AI models. Microsoft's AI runs faster and smoother on this platform, and you can get output with a single click. Therefore, this demonstration uses it.
The demonstration also includes two AI tools to colorize and enhance the recovered photos. Both are available for free on his Replicate and Google Colab. We encourage you to give it a try, but Microsoft's AI is sufficient for basic restores.
How to restore, colorize and upscale old photos
Handing old, degraded photos to Microsoft's AI is just the first step in my process. You can then use AI to colorize the recovered photo (this part is optional). Finally, a third AI is used to upscale the entire image.
All you need to get started with Replicate is a GitHub account. If you already have GitHub, skip this step. If not, please proceed to GitHub. You need a valid email address to join GitHub, but signing up only takes a few minutes.
old photo restoration
Once you've logged in to Replicate, you're ready for the first step.In the search bar, type bring old photos back to lifeor go directly to the page.
Click on the little cloud icon to select and upload the photos you want to recover.
For high-resolution scans and if the photo has scratches, check the relevant boxes. Click Run. The output takes approximately 2 minutes to process. This tool has a handy slider to show before and after.
Notice the Warm or Cold tag under the AI model title. A cold tag means the model is starting up and the photos will take longer to process. If the model is already warmed, you will get the output faster.
As you can see, the AI did a good job of cleaning up the photo. Click “Download” to get the recovered image in full resolution. Or click Tweak to run the AI again with different settings.
Colorize restored photos
Add color to the recovered photo before upscaling it. You can skip this step if you want to keep the original color.
Use Deoldify AI. We tested nearly half a dozen other models on Replicate that were built for the same job. But Deoldify consistently produced the best results.
To run Deoldify on the restored image, run the following search: become antiquated Click on the Replicate search bar or go directly to the Deoldify page.
Click the upload icon and select the recovered file you just downloaded.
Deoldify has two settings: model name and rendering factor. Try both until you get the best results. The default is Artistic Model, with a rendering factor of 35, which works in most scenarios. I had better results using Stable here.
Once you have selected your preferred settings, click Run. Wait for processed output. Once you are satisfied with the results, click Download. Or click Tweak to reprocess with different settings.
upscaling old photos
So far, I have restored severely damaged photos and added color to them. To conclude this restoration, we will upscale the processed photo. Upscaling uses AI to add pixels to an image, making it bigger, sharper, and sharper. Remember the “zoom to strengthen” metaphor from CSI?
Open ESRGAN in Replicate. Click the upload button and drop the colorized image as input.
Next, enter the scale. Scale here refers to image size. I set this to 4. This will result in an image that is 4 times larger than the size you entered. So for a 1080P image, you will get 4K as output. You can also select Face Enhancement to sharpen facial details. Finally, click Run.
Once completed, the before and after previews will look the same. However, when you download the processed file, the resolution will be much higher.
That's all. The restored photos are now ready to be used in a digital or print album. Let's compare them side by side.
Including these three AI models, Replicate has a whole collection of image restoration models. There is also a catalog dedicated to image upscaling and model adjustments.
One in particular that caught my eye was Google's Maxim AI. You can blur, remove noise, remove contours, and remove haze on your photos with just one click. The results were amazing. It works wonders with photos taken with Polaroid or low-resolution Danphone cameras. You can even use Google's AI magic to retouch old photos taken in low light.
You can restore and enhance most of your old photos in three easy steps. However, you may need to play around with the settings to get the perfect output. We also encourage you to try other repair models available for free on Replicate. You can modify my workflow or create your own.
