2x vs 4x vs 8x vs 16x: Which Upscale Factor Should You Choose?
The factor only decides one thing: how many pixels come out. Here is how to work backwards from your target, and when a higher factor makes the result worse instead of bigger.
The multiplier is the least interesting setting on the page and the one people agonise over most. It does exactly one thing: it decides how many pixels come out. Everything else — sharpness, invented detail, processing time — follows from that and from the model.
Work backwards from the output size
Decide what the file has to be, then pick the smallest factor that gets there. A 2x step on a 1000-pixel image gives 2000 pixels; a 4x step gives 4000. Nothing else changes.
| Source | 2x | 4x | 8x | 16x |
|---|---|---|---|---|
| 300×300 sprite | 600 | 1200 | 2400 | 4800 |
| 800×600 old photo | 1600 | 3200 | 6400 | 12800 |
| 1920×1080 wallpaper | 3840 (4K) | 7680 (8K) | 15360 | 30720 |
| 1200×1600 phone portrait | 2400 | 4800 | 9600 | 19200 |
Two things fall out of that table. First, most everyday targets are reached at 2x or 4x: 4K from a 1920 wallpaper is exactly 2x, an A4 print at 300 DPI from a phone photo is about 2x. Second, past 8x you are usually producing dimensions that no screen and no consumer printer will ever ask for.
When 2x is the right answer
- The source is already reasonable — a phone photo, a 1920 wallpaper, a modern game screenshot.
- The output is for a screen or a photo print up to A4.
- You care about the result looking like the original, only bigger. Two-times upscaling is the range where reconstruction is almost invisible.
When 4x earns its place
- A small source with a high target: a 900-pixel scan that has to become an A3 print, a 400-pixel avatar that goes on a poster.
- Line art and illustrations, which survive higher factors better than photographs because there is less texture to invent.
- You need cropping headroom after the upscale.
8x: for genuinely tiny sources
8x makes sense when the input is very small and the output needs to be normal-sized: a 300-pixel sprite for a 2400-pixel asset, a stamp-sized scan, an icon from an old app. On an already decent photo it mostly buys processing time and detail that was not in the file — zoom to 100 % on an 8x portrait and you will see pores the camera never recorded.
16x: read this before using it
16x exists because sometimes it is the only way to reach a size, not because it is a quality setting. On a 300×300 sprite it is useful. On a 1920×1080 wallpaper it produces 30720 pixels wide, four times the file size, and a result that mostly shows what the model likes to draw. If you are considering 16x, the honest alternative is to find a better source first.
What actually degrades as the factor grows
- Invented texture. Every step reconstructs more than it recovers. At 2x you are sharpening; at 16x you are largely inventing.
- Time and quota. Cost scales with output pixels, so 16x is sixteen times the work of 4x on the same input.
- Hallucinated structure. Faces and lettering are where high factors fail most visibly: the model produces a plausible face, not the actual person, and reshapes letters it does not recognise.
A three-step decision
- Write down the target in pixels (or in centimetres at 300 DPI).
- Divide by the source size. That number is your factor — round up to the next available step.
- If the answer is 8x or 16x, ask whether a better source exists, or whether the target can be smaller. Both usually work better than pushing the multiplier.
Nothing here is a secret setting: enlarge an image free up to 3000×3000 pixels without an account and compare 2x against 4x on the same file — the difference is usually obvious within a minute. For the settings that go with the factor, see the guide to enlarging without losing quality.