No. 07 — AI Tools · Cloud HDPro · on our GPU

Photo quality, improved for real.

Quality is five separate things, and a model can move three of them. Knowing which attribute is failing in your photograph is the difference between a Cloud HD run that visibly pays and one that quietly does not — so this page names all five before it offers you the button.

Pro · Cloud HDDeleted on completionUp to 25 MB · 24 MP
Improve a photoCloud HD · our GPU

How it works

  1. Pick a photo whose fault is sharpness, noise or faded colour — JPEG, PNG or WebP up to 25 MB and 24 megapixels.
  2. It is uploaded to our server for that run and deleted when the run finishes, and in all cases within 24 hours.
  3. Download it at the same size, cleaner and sharper. One general Cloud HD run out of your monthly allowance.

This page is the AI photo enhancer, tuned to deciding which photographs deserve a run.

All Image Tools

AI Tools

Solutions by use case

Improve Photo QualityPro, on our server

Break quality into attributes and the vague complaint becomes a decision. Pixel count is how much picture the file physically holds. Sharpness is how well fine structure survived focus, shutter and lens. Noise is the random speckle laid over everything. Tonal range is how much separation survives in the brightest and darkest areas. Colour accuracy is whether white looks white. Of those, the Cloud HD enhancer moves sharpness and noise a long way, colour a little, and pixel count not at all — it returns exactly the dimensions it was handed. Clipped highlights are gone permanently, whatever any tool claims.

Choosing what not to run is half the skill. The monthly allowance — 150 general runs on Pro Monthly, 200 on Pro Annual, resetting on the 1st in UTC with no rollover — rewards selectivity, and so does the result. A correctly exposed, well-focused photograph gives a reconstruction model very little to reconstruct, so the output looks almost identical and the run is spent. Save them for the photographs that are visibly soft, speckled, over-compressed or faded, where there is real damage to undo. The related page on enhancing photo quality goes into what enhancement can and cannot recover on a single image.

Some of the biggest gains cost nothing at all. Stop re-saving JPEGs: every round trip discards a little more, and five forwards through a chat app can do more damage than the original mistake. Export once, at high quality, and use the image converter when a format change is genuinely needed and the image compressor once, at the end, for delivery. Both run entirely in your browser. If what you actually need is a bigger file rather than a cleaner one, that is the pixel-count attribute and it belongs to the upscaler.

Where the file goes when you do decide a run is worth it. Almost everything on MiniPx processes your image inside your own browser and uploads nothing. Cloud HD is the deliberate exception, because these models are far too large to ship to a tab: it is Pro-only and opt-in, your photo is uploaded to our server for that run and deleted when the run finishes, and in all cases within 24 hours, it is processed in the United States, encrypted in transit, and never used to train anything. The model behind the run is RestoreFormer++, and the result says so. Prices exclude tax and a refused or failed run is never charged.

How it works

  1. Name the fault before you spend anything: Decide which of the five attributes is actually failing. A run aimed at the wrong attribute is the usual reason a result underwhelms.
  2. Shortlist the photos worth a run: Soft, noisy, compressed or faded images gain the most. Already-sharp images gain little and are not worth the allowance.
  3. Upload the original, not a copy of a copy: JPEG, PNG or WebP up to 25 MB and 24 megapixels. Every intermediate save has cost the model something to work with.
  4. Compare crops at actual pixels: One textured area, one smooth gradient, both at 100%. That comparison answers the question a fitted-to-screen view cannot.

Frequently asked questions

Which measurable parts of quality can actually be moved?
Five attributes describe most of what people mean by quality: pixel count, sharpness, noise, tonal range and colour accuracy. A Cloud HD run moves sharpness and noise substantially and colour modestly. It does not change pixel count — that is the upscaler’s job — and it does not recover highlights that clipped to pure white or shadows that clipped to pure black.
I have forty photos from an event. Should I run them all?
No, and the allowance is a good discipline. Sort by whether the photo has a problem a model can move: soft, noisy, compressed or faded gets a run; correctly exposed and sharp does not, because there is little for the model to reconstruct. Twelve deliberate runs beat forty hopeful ones, and you keep the rest of your monthly allowance.
Does saving as PNG instead of JPEG improve quality?
It stops further loss; it does not undo loss already baked in. PNG is lossless, so re-saving a damaged JPEG as PNG preserves the damage perfectly at a much larger file size. Convert to PNG when you are about to edit repeatedly, and use the on-device converter and compressor for the final delivery format.
Would a higher megapixel camera have made these photos better?
Only for the pixel-count attribute, and usually less than people expect. Sharpness is set by focus and shutter speed, noise by sensor size and light, tonal range by exposure. A 48-megapixel phone photo taken in a dark restaurant is a large file with all the same faults. Fixing the capture conditions beats buying resolution.
How do I tell whether the processed version is genuinely better?
Compare the same crop at 100% rather than the whole frame fitted to screen — fitted views hide both the improvement and the artefacts. Look at one region with fine texture and one with a smooth gradient. If the texture gained and the gradient stayed clean, keep it. Keep the original either way; it costs nothing.