Member Avatar for Reverend Jim

I don't often rave about software (ok, other than my own), but I have to give two big thumbs up to the latest release by Topaz Photo AI.

I have a very large collection of family photos reaching back to the turn of the last century. I long ago converted them to digital but I have always been disappointed that so many of them were fuzzy. Through one of my many newsletters I came across the latest release of Topaz Photo AI. While there are many packages that allow tweaking of digital images, I have yet to see anything that is so easy to use (with default settings) while producing outstanding results.

The package, which retails for just under $200, includes three modules that if purchased separately would cost around $250. The modules are

  1. gigapixel (for upsizing)
  2. denoise (noise removal)
  3. enhance (facial recovery and sharpening)

The software, as downloaded, is fully functional, lacking only the ability to save the results. If you want to save your work you have to pay. Sort of. When you drop an image onto the interface and scroll via mouse-wheel until the image is fully displayed, the AI engine immediately goes to work analyzing the image. Depending on the complexity and size of the image and the capabilities of your system, the processing could take up to 30 seconds. Photo AI offloads as much processing as possible onto your GPU so a good graphics card really helps.

After the processing is complete, the displayed image is replaced by the enhanced image. Normally you would do a save at this point, but if you are happy with the result as displayed you are free to do a screen capture and save the new image that way. My screen resolution is 1920x1080 and the enhanced image is displayed in a panel approximately 1600x900. Most of my scanned photos are smaller than 1600x900 so a screen cap is easily sufficient for my needs.

Emphasis is placed on recovering faces and I have noticed that in extremely degraded images, faces are recovered quite well while clothing is not, however even with this, the resulting image is still better than the original. Automatic noise reduction is also excellent, although in some cases I still require manual touch-ups. I use FastStone Viewer (free) and I find that I can do all manual adjustments from within that application (contrast, brightness, gamma, clone/heal, etc.).

I have also found that even when taking an image which is only slightly blurry, enhancing, then flipping between the original and enhanced images demonstrates just how much nicer the slightly sharper image is.

The install will take approximately 1.2 GB of drive space.

before_1.jpg

after_1.jpg

before_2.jpg

after_2.jpg

before_3.jpg

after_3.jpg

  • 2 Contributors
  • 1 Reply
  • 191 Views
  • 3 Years Discussion Span
  • Latest Post Latest Post by TonyEveret
pritaeas commented: Those are some impressive results +17
tarael1 commented: Remarkable. Thank you for sharing. +2

Dani AI

Generated

Good follow-up to ’s image examples. The dramatic before/after makes one thing clear: good results come from a repeatable workflow, not just a single pass through an enhancer. The checklist below focuses on steps you can apply immediately to get consistent, archival-quality restorations while avoiding common AI artifacts.

  • Archive a lossless master first. Keep the untouched scan and work on a copy so nothing is lost.
  • Scan at the highest true optical resolution your scanner provides and save in a lossless format. Capture the full tonal range; avoid in-scanner sharpening.
  • Do light pre-cleaning: crop, deskew, and remove obvious dust/scratches in an editor before running an automated enhancer. Small manual fixes prevent the AI from amplifying damage.
  • Test on a small crop. Run your preferred enhancement on a sample region (a face and a patterned area) to check how the tool treats texture, edges, and text.
  • Use a two-stage approach: reduce noise first, then recover detail/upscale. When available, apply masks so delicate areas (faces, printed text) don’t get over-processed.
  • Finalize manually. After enhancement, do a careful clone/heal pass and subtle tone and color corrections. Save two outputs: a lossless archival file and a web-friendly copy with embedded color profile.

Watch for hallucinations (invented details), edge halos, and over-smoothed skin. Inspect at 100% and at larger zooms. For large batches, automate settings but spot-check samples from different scan conditions. Keep a short log of settings used so you can reproduce results later. The before/after images posted by illustrate what’s possible when careful scanning, modest manual cleanup, and cautious AI enhancement are combined.

Member Avatar for TonyEveret

Those are some really impressive results. Thanks for sharing such motivating progress.

Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.