The order matters
Enlarging a blocky picture enlarges its blocks; denoising after enlarging smears grain into blobs. So the chain runs in the order that keeps each step's input clean: blocks are smoothed on the file's own grid first, light and colour are then measured and corrected on the cleaned picture, a light sharpen is set from the measured edges, and the enlargement runs last, when the file is saved. The receipt prints the stages in that order.
Reading the damage
| Reading | Comes from | Decides |
|---|---|---|
| Block ratio | steps across 8×8 boundaries vs everywhere else | whether deblocking is worth it |
| JPEG q≈ | the luminance quantisation table in the file | how strong the deblock is |
| Grid offset | the boundary column with the largest step | where the filter is aligned |
| Long edge | the file | the size step |
The quality estimate is a reading of the file, not a guess about the scene: JPEG writes its quantisation tables into its header, and the standard IJG scaling can be inverted to recover the quality setting that produced them. Files with custom tables are reported as approximate.
When to use the local model
If the classical 2× result is clean but soft, the Local AI panel offers models that draw sharper texture. Download size and an estimated time are shown before anything is fetched, a region preview comes first, and the difference view shows which parts of the picture the model changed and by how much.
Limits
- Compression discards detail permanently; deblocking removes the squares, not the loss.
- Text in a heavily compressed screenshot may stay soft; the text image route has a pixel-exact enlargement for that.
Questions people ask
- What makes a picture look pixelated?
- Usually two things at once: it is small, so each pixel is large on screen, and it has been JPEG-compressed hard, so it is made of visible 8×8 squares. Forwarding a picture through a chat app often does both. This page treats them in that order: blocks first, then size.
- Can it turn a low quality image into a high quality one?
- It can make it cleaner and larger. It cannot bring back the detail the compression threw away. Deblocking smooths the squares where the step between them is small enough to be compression; enlarging gives you more pixels to look at; a local model, if you choose one, draws plausible texture and is labelled reconstructed.
- Why is the deblock off on the preview?
- When the picture is bigger than the 2,048 px working copy, the copy is resampled and its 8×8 grid no longer lines up with the file's. The deblock then runs on the full-resolution picture for the loupe and the saved file only, and the panel says so.
- Is anything uploaded?
- No. Every step runs in this tab. A local model, if you choose one, is downloaded to your device and runs there.
- Can I switch off one step?
- Yes. Deblocking and the other clean-up stages are switches in the Clean-up panel, the size is in the Size panel, and every stage of the measured pass is a chip you can mute. The receipt lists exactly the steps that ran.
Related
- JPG quality enhancerReads the file's own compression tables, then cleans the 8×8 blocks at the same size.
- HD photo enhancerHD, Full HD or 4K long edge, classical or local model, with the exact factor printed.
- AI photo enhancer, on your deviceThe measured pass first, then local models for 2× detail — download size shown before anything is fetched.
- JPEG quality and the 8×8 blocksWhat the quality number in a JPEG means, how it can be read from the file, and what deblocking can undo.
