Try it yourself: UpgoLAB runs in the browser and takes a 720p clip to 1080p or 4K without installing anything. Free, no watermark.
The numbers in this post are from my own test clips, not marketing material.
Most people check the result on a phone screen, decide it looks fine, and send it to print. The disappointment arrives later: plastic skin, fabric turned into a flat patch, a grey rim around every object.
The cause is almost never the intensity setting. It is where and how you check.
Why the full-screen view lies
Your eye cannot separate resolution from detail. A larger image always reads as sharper at a glance, because more signal reaches the retina.
At 100% that advantage disappears: you see one screen pixel per image pixel, and the estimation errors become visible.
The only valid test is: same crop, same zoom, side by side, at 100%. Anything else measures file size, not quality.
The 3 points worth checking
Only three. You do not need to scan the whole frame.
1. Hair
It must separate into strands, not merge into a mass. A continuous surface where hair used to be means the preset was too aggressive.
2. Fabric weave
The pattern must keep its periodicity. If the fabric became a uniform surface, the model interpreted the texture as noise and removed it.

The pattern must stay regular, with no stretched zones or flat areas.
3. Object edge against the background
It must be clean, without a grey or white halo. A halo almost always comes from sharpening before scaling, not from the preset.

Over-sharpening creates a white halo on the edges.
The trap: judging by size
If a face is 40 px wide in the image and 200 px on screen, you are looking at the browser's interpolation, not at your result.
Always compare at the same physical scale: the same zoom percentage on both images.
The most reliable signal: skin
For portraits, the criterion is not sharpness. It is whether the eyes and lashes stay separated.
If the skin is smooth like plastic, shiny and uniform, the mode was wrong. Lowering the intensity will not fix it: the mode has to change.

Preset PORTRAIT STUDIO. Left the original, right the result. Skin texture survives.
Before checking, verify the order
If the processing order was wrong, checking is pointless because the damage is already done:
- Denoise first. Compression noise is made of 8x8 pixel blocks. Upscale first and the model amplifies it, after which it cannot be removed.
- Upscale at most 2x. From 1000 to 2000 pixels, not 4000.
- Sharpen last, lightly.

Preset SOCIAL DENOISE. The blocks disappear, the shape remains.
Where I test it
I use UpgoLAB in the browser. Modes are separated by material type, and the denoise, upscale, sharpen order is already wired internally — which is exactly the part most people get wrong by hand.
There is a guide to choosing the preset per photo type. Free, no install.
It is also available in the blog, in English.
What does not work
- Information that was never captured does not come back. Fully blurred text may become legible, but not correct.
- Motion blur is not recoverable. Direction information is lost at capture time.
- Material compressed several times has a low ceiling. Fine detail is already gone.
Summary
- Check at 100%, never in full-screen view
- Only 3 points: hair, fabric weave, edge against background
- Plastic skin means wrong mode, not high intensity
- If the order was wrong, checking is pointless
Images in this article are comparison examples generated with UpgoLAB.
Measured write-ups behind these numbers, with the browser tests reproduced: AI upscaling notes.
Further reading, all measured the same way: upgolab-old-footage-restoration; stretching a single axis: anisotropic upscale, measured; upgolab-tools; halftone dots under upscaling: the lattice scales, descreening costs 1.2x of it.
The orientation trap in point 1, measured on all eight EXIF values: EXIF orientation and upscale order.
Top comments (0)