Image Upscaling Guide
How AI image upscaling works, when to use it, and how to get sharp, print-ready results from small photos.
Key takeaways
- AI upscaling reconstructs detail rather than just stretching pixels.
- Best for small, old or compressed photos you need larger or sharper.
- Pair it with enhancement or restoration for the cleanest result.
- Match the upscale factor to your target dimensions and clean compressed files before enlarging them.
What upscaling actually does
Traditional resizing stretches existing pixels, which looks blocky. AI upscaling instead predicts the detail that would be there - edges, texture, fine lines - and reconstructs a larger, sharper image. The Image Upscaler does this in one step.
When to use it
Reach for upscaling when a photo is too small to print, when a thumbnail is the only copy you have, or when a compressed download looks soft. It is also a great finishing step after a face swap to make the result print-ready.
What it can and cannot do
Upscaling adds plausible detail, but it cannot invent information that was never captured - a heavily blurred face will improve, not become a perfect portrait. Start with the best original you have for the best outcome.
Combine for best results
For old or damaged photos, restore first, then upscale. For dull or low-contrast images, run the enhancer after upscaling to refine light and colour.
Understanding upscale factors and final dimensions
Upscaling is described as a multiplier - 2x doubles each side, so a 1000x1000 image becomes 2000x2000 and contains four times the pixels, while 4x produces sixteen times as many. To work out the factor you need, divide your target dimensions by the original: a 600px-wide scan printed at 2400px wide needs a 4x pass. A few rules of thumb:
- Web and social: 2x is usually plenty
- A4 / photo prints: aim for roughly 300 DPI, often 3x-4x from a phone photo
- Large posters: upscale in stages rather than one extreme jump
Why compression artefacts confuse upscalers
Heavily saved JPEGs contain blocky 8x8 compression artefacts and colour banding that the model can mistake for real detail - upscaling then sharpens the flaws along with the subject. The fix is to clean the image before enlarging it: a pass through the Image Enhancer reduces noise and smooths banding so the upscaler has a truer signal to reconstruct from. Screenshots of screenshots, social-media re-saves, and messaging-app downloads are the usual culprits, since each re-save compounds the damage. Where you have a choice, always source the original file rather than a forwarded copy. For old prints with physical damage, restore first - see the photo restoration guide - then upscale.