Old photos often need more than a basic resize. AI upscaling helps by rebuilding edges and texture so a small or soft image can hold up better on modern screens.
If you are starting with scans, family snapshots, or small legacy files, the goal is not to invent detail that never existed. The goal is to preserve structure, improve clarity, and export a cleaner version that looks better at the size you actually need.
Use the highest-quality copy available before you upscale anything. If you have both a compressed social-media image and an older direct export or scan, use the direct export or scan. Reducing compression damage before upscaling gives the model more usable information.
It is tempting to jump straight to 8×, but that is not always the best option. If you only need enough resolution for a web post or a small print mockup, start with 2× or 4× and review the result. Over-upscaling can make faces, hair, and texture look unnatural.
Best for modest improvements when the source image already has usable detail.
A good default for small photos that need a clear improvement on modern displays.
Portraits and older scans can look good at first glance but still break down around eyes, hair, clothing texture, and high-contrast edges. Review those areas before you export.
If you know the exact output size you need, upscale first and then resize to the final target. That workflow usually produces cleaner results than stretching a small image directly to the destination dimensions in one step.
Use the target-resolution mode when you know the minimum size you need but do not want to overshoot by more than necessary.
If the image starts looking brittle, over-sharpened, or obviously synthetic, step back to a lower scale factor. A believable 2× or 4× result is usually more useful than a larger file with distracting artifacts.