Restore First or Colorize First? The Correct Photo Restoration Order

You have a scratched, faded black-and-white photo from 1952. You have three AI tools: a restorer, a colorizer, and an upscaler. There are six possible orders to run them. Five produce garbage. Only one consistently works.

I tested all six sequences on ten damaged vintage photos — torn edges, water stains, fading, dust spots, the works. Here is what happens with each order and why only one makes sense.

The six orders, tested

Restore → Colorize → Upscale (Winner)

Fix the damage first on the clean black-and-white original. The AI restorer has an easier job because it is working with monochrome data — scratches and dust stand out clearly. Then colorize the restored image. The colorizer gets a clean source and produces accurate colors. Finally, upscale the finished color image. Result: sharp, clean, natural-looking color photo. This order worked perfectly on 9 out of 10 test photos.

Restore → Upscale → Colorize (Second place)

Restore first, then upscale the black-and-white version, then colorize. This works but the colorizer sometimes produces slightly oversaturated colors on upscaled images because the upscaler adds synthetic detail that the colorizer misinterprets. Result: usable but colors need manual toning down on 3 out of 10 photos.

Colorize → Restore → Upscale (Fails)

This is the most common mistake. People want to "see the colors first" so they colorize immediately. Big problem: the restorer now has to fix scratches on a color image, which is harder — scratches that were obvious on black and white blend into color gradients. The restorer also sometimes "fixes" intentional color variations, creating blotchy patches. Result: muddy colors with visible scratch artifacts on 7 out of 10 photos.

Colorize → Upscale → Restore (Fails badly)

Colorizing and upscaling first locks in the damage. The restorer at the end cannot distinguish between original scratches and upscaling artifacts. Result: unusable on 8 out of 10 photos.

Upscale → Restore → Colorize (Fails)

Upscaling first amplifies every scratch and dust spot. The restorer then has to fix damage that is now 4x larger and more detailed. It overcorrects and creates smooth patches that look obviously edited. Result: over-processed, waxy-looking faces on 6 out of 10 photos.

Upscale → Colorize → Restore (Fails worst)

Everything that can go wrong goes wrong. Amplified damage confuses the colorizer, which assigns wrong colors to scratch artifacts, which the restorer then tries to fix and makes worse. Result: completely unusable on 9 out of 10 photos.

Why Restore → Colorize → Upscale works

The principle is simple: each step should make the next step's job easier, not harder.

Restoration removes damage. That makes the image cleaner, which helps the colorizer assign accurate colors — it is looking at actual image content, not scratch patterns. A clean colorized image then upscales cleanly because the upscaler is enhancing real detail, not artifacts.

The counter-intuitive part: do not upscale before colorizing. Common sense says "make it bigger so there is more detail to work with." But upscaling adds synthetic detail — pixels the AI guessed. The colorizer then colors those guessed pixels, and small errors compound. Keep the image at its original resolution through restoration and colorization. Upscale last.

The face-specific exception

For photos where faces are the main subject and are severely damaged, use Face Pro mode in the photo restorer. It uses GFPGAN specifically trained on facial restoration. Run Face Pro first on the original, then the standard Auto mode for the rest of the image, then colorize, then upscale. Face-first restoration prevents the "waxy skin" look that happens when general-purpose restoration smooths facial details too aggressively.

Ready to restore your old photos in the correct order? Create a free account and get 5 credits. The full pipeline (restore + colorize + upscale) costs 9 credits total. For a deeper dive into the restoration process, here is the complete guide to restoring old photos with AI.