After the Metadata: AI Denoise and Upscale Options

ExifToolGui’s job ends where the pixels begin. It inspects and fixes what a photo says — dates, credits, keywords, location — and it does not denoise, sharpen or enlarge the image itself. For the frames that need that work, the current tool class is AI photo editors: programs that use learned models to denoise high-ISO shots, upscale low-resolution files and generally rescue detail. This page covers the options linked from this site, described only at the level that can be stated responsibly — what each is for — with no invented version numbers and no prices. It closes the Workflows & Buying Guides hub sequence that starts with the metadata-first Lightroom workflow and the backup workflow. The metadata tools are on the download page.

Why the metadata pass comes first

AI editing produces new files, and new files are where metadata goes to be rewritten, dropped or quietly altered by export settings. Run the order this site teaches — metadata first — and two good things happen. Your masters carry true dates and credits before any editor touches them, so derivatives start from the truth; and you build the checking habit that catches what an export changed, by re-reading the finished file in ExifToolGui afterwards. The workflow is a loop, not a line: fix, edit, re-check. Skipping the first fix means every upscale of a family scan carries the scanner’s wrong date into the future, beautifully denoised and permanently mislabelled.

The options, described plainly

• Topaz Photo — an AI photo editor centred on denoise, sharpening and upscaling problem frames: the low-light shot, the heavy crop, the small old file that needs to become printable. Linked here as Topaz Photo.
• DxO PhotoLab — a RAW editor whose AI denoise is its headline strength, used by photographers who want the noise work inside a full develop workflow rather than as a separate rescue step. Linked here as DxO PhotoLab.
• Luminar Neo — an AI-driven photo editor positioned on creative editing with automated assistance, the option readers who dislike subscription-style suites often ask about. Linked here as Luminar Neo.

Some links on this page are affiliate links — see our affiliate disclosure. Those three descriptions are deliberately category-level. Features, versions and prices change faster than any guide can responsibly track, and the vendors’ own pages — reached through the links — are the place for the current specifics and trial terms.

Choosing by the problem in front of you

Noise is the problem (high-ISO events, old compact-camera files, night scans): denoise-led tools are the shortlist, and the honest test is your own worst file run through a trial, judged at the size you will actually print or share. Resolution is the problem (small originals, deep crops, phone photos destined for a wall): upscaling is the headline feature to test, again on your file, because source quality decides what any upscaler can honestly invent. The whole edit is the problem (you want developing and AI assistance in one place): a full editor with AI features fits better than a single-purpose rescuer. In all three cases the discipline is identical: work on a copy, export deliberately, and re-check the result’s metadata before it joins the library — dates and credits verified in ExifToolGui, location present or absent by your decision, not by an export default.

The archival case deserves its own sentence, because it is where this tool class earns the most gratitude. Old family scans and early digital files arrive with everything stacked against them: noise, low resolution, faded contrast — and, very often, scanner-written dates standing where capture dates should be. The full treatment is the loop in miniature: fix the dates on the scans first (our scanned-photo date guide covers setting a real capture date), denoise and upscale the copies, then verify that the finished files still carry the corrected dates and the family names you keyworded. Sentimental value is still value, and it deserves accurate records as much as client work does.

What AI editing cannot do

It cannot recover information that was never captured — upscaling synthesises plausible detail, it does not retrieve the real detail a bigger sensor would have recorded, and denoising trades texture for smoothness in ways that deserve a zoomed look before you commit a whole set. It cannot fix wrong metadata; a denoised file with the wrong date is still wrong, which is why this page is sequenced after the metadata guides rather than instead of them. And it cannot substitute for the archive habits in the backup workflow: the rescued frame, once you love it, needs the same three-copy protection as everything else you keep.

Frequently Asked Questions

Should AI editing come before or after metadata work?

After. Fix dates, credits and keywords on the masters in ExifToolGui first, edit copies in the AI tool, then re-read the exported file’s metadata and repair anything the export changed. That loop is the whole method.

Which of the three options is best?

The one that handles your actual problem file best in a trial: denoise-led rescue, RAW development with AI denoise, or AI-assisted creative editing are different centres of gravity. This page describes categories honestly and links the vendors; it does not crown a winner without your test file.

Will an AI editor preserve my metadata?

Not automatically — export settings decide what is written, kept or dropped. Verify with ExifToolGui after export: capture date, Artist, Copyright, keywords and GPS state, on every file that matters.

That completes the desk: metadata fixed, library backed up, problem frames rescued. Browse the full sequence in the Workflows hub, and revisit the comparisons in the Comparisons hub when the next tool decision arrives.