Detail inspector
Image Upscaler Hub
An upscale is judged at 1:1 or it is not judged at all. Fitted to a screen, almost every result looks like an improvement; at actual pixels you can see which details were recovered and which were invented. This site works at that magnification.
- Subject
- image upscaling
- Viewed at
- 1:1, always
- Status
- no reviews yet
- Scoring
- none — words only
Inspection pass
Six places to put the loupe
These are the regions where an upscale gives itself away. They hold for any tool and any image, which is why they are worth knowing before you have picked either.
Fine dark strands against skin. Look for lashes that have merged into a solid band, or a hairline that has acquired a clean edge it never had.
Pores, fine lines, the slight unevenness of a real face. If these have been replaced by a smooth waxy surface, texture was discarded rather than resolved.
The only region where the answer is checkable. Letters that sharpened into different letters are the clearest evidence a model is guessing.
A roofline, a branch, a shoulder. A bright or dark fringe tracing the edge is over-sharpening rather than recovered detail.
Brick, mesh, woven fabric. Models tend to tidy these into something more regular than the original, which reads as impressive and is a fabrication.
Large images are often processed in sections and stitched. Look across the middle of a flat area for a faint seam or a change in texture character.
Scale
What the multiplier costs
2×
The conservative setting. There is usually enough information in the source for the result to be mostly interpolation, so there is less for a model to invent and less that can go wrong.
4×
The common default, and the point where the output becomes mostly prediction. One source pixel is now sixteen, and fifteen of them are the model's opinion.
8× and beyond
Often achieved by running a 2× or 4× pass more than once, which compounds whatever the first pass invented. Results can look striking and bear a loose relationship to the original.
Does the source format matter?
Considerably. JPEG compression leaves blocky artefacts that an upscaler will faithfully enlarge, and sometimes mistake for structure. A heavily compressed source is a harder problem than a merely small one.
House rules
How this will be written
No invented numbers
No scores out of ten, no star ratings, no tier boards. Where something is better, it gets said in words and the reason is shown.
Only claims we can stand behind
Nothing describes a test that did not happen. Pricing and feature claims are checked against the vendor's own pages.
No paid placements
Nothing here is sponsored, and outbound links carry no affiliate tracking and earn no commission. No company pays to be included or to be written about more kindly.
This site is new and there are no tool write-ups on it yet. The crop numbers above are a checklist of where to look, not results from an inspection we have run. When write-ups land, each one will say which crop it is showing and at what magnification.