Images & photos · Image Converter & Compressor
How to check a compressed image for artefacts before you publish it
· Why it matters
image-quality image-compression publishing
A smaller file is only a win if nobody can see the damage, and thumbnails hide almost everything. This post describes a short inspection routine for compressed images and the artefacts each format tends to produce.
The banner that looked fine until someone zoomed in — the problem with judging compression from a preview
A banner may look acceptable while fitted into a narrow preview and reveal halos or coarse patches when opened at full resolution. Preview scaling blends neighbouring pixels and can hide damage. ToolAcre’s first-result pane is useful for orientation, but editorial approval should include the downloaded file in a viewer where zoom and interpolation are understood.
Do not wait until the asset is in production to discover the issue. Define an inspection step beside the byte budget and accessibility check. The person approving quality should know the intended display sizes, because an artefact invisible in a thumbnail may dominate a cropped desktop hero or a later social reuse.
View at 100 percent, not 'fit to screen' — why scaled previews smooth over blocking and ringing
At 100 percent, one image pixel maps to one display pixel in a conventional viewer. That removes fit-to-window smoothing from the first pass and makes adjacent candidates easier to compare. High-density displays and browser zoom can complicate the literal mapping, so record the viewer and use the same setup for original and outputs.
Native-size review is not the only view that matters. Follow it with the actual layout because resampling at delivery can expose or conceal different details. The two checks answer different questions: what did encoding do to the file, and what will a reader see after the page renders it?
Where to look first — high-contrast edges, smooth gradients, skin, sky and saturated text
Inspect high-contrast edges, smooth gradients, skin texture, hair, foliage, shadows and saturated text. Each region stresses the encoder differently. A single central crop can miss a damaged logo in the corner or banding across a sky. Use a short repeatable route through the image so reviewers examine equivalent areas.
Include transparency and matte edges where applicable. JPEG will flatten alpha to the selected background; WebP and PNG can retain it under ToolAcre’s format model. A clean photographic center does not excuse a fringe around the product cut-out. Approval criteria should follow the image’s purpose, not whichever area happens to look best.
Recognising each artefact — blocking, ringing, banding and colour bleed, and which format and setting causes each
Visible symptoms may include block-like regions, ringing near sharp boundaries, stepped gradients or colour bleeding around saturated details. The browser encoder is a black box to this application, so report what you observe rather than claiming a particular hidden table or subsampling mode caused it. The file, settings and crop provide reproducible evidence.
Also distinguish resampling blur from encoding artefacts. Keep dimensions constant during the first quality comparison. If softness appears only after resizing, geometry is the relevant variable. If it grows as lossy quality decreases with fixed dimensions, encoder settings are implicated. Controlled comparisons prevent every flaw from being mislabeled “compression.”
Name observed artefacts without assigning undocumented encoder internals or fixed format causes
Flip between original and candidate at the same zoom and viewport position. Avoid side-by-side panes with different scaling, because a smaller display can look cleaner merely by hiding detail. ToolAcre reports measured bytes, but the lower number is context, not a verdict. A file that fails inspection is too expensive in quality regardless of savings.
Keep notes on where a difference first became objectionable. “Roof edge at upper right rings” is actionable; “looks bad” is hard to reproduce. A second reviewer can then check the same location and decide whether the issue survives the intended display size. This converts taste into a bounded editorial decision without inventing an objective score.
Worked example: a product photo compressed at three qualities — finding the point where artefacts appear and stepping back one notch
Choose one representative product photograph and export three candidates from the untouched original at clearly separated quality settings. Do not use one candidate as the source for the next. Hold format and dimensions constant, name files with their settings and record actual bytes from each conversion.
Inspect the planned regions at 100 percent, beginning with the most compressed candidate. When damage becomes unacceptable, move to the next less aggressive file and repeat. Approve the first candidate that passes every critical region, then verify it in the site component. No universal quality number belongs in this method because image content and browser output differ.
Worked method with three settings and no invented quality threshold
Automated metrics can support large pipelines, but ToolAcre calculates none. It also cannot compensate for monitor calibration, display sharpening or a content management system that later recompresses the upload. Those variables belong in broader publishing QA and should not be hidden behind local approval.
This inspection routine is intentionally human and file-specific. It catches obvious damage before publication and creates a defensible choice among local candidates. For regulated imaging or pixel-level analysis, use domain-specific validation rather than an editorial visual check.
Takeaway: inspect, then ship — how to iterate quickly in the Image Converter & Compressor because nothing is uploaded and every attempt is local
Inspect, then ship. The converter shortens iteration because every attempt is made in the tab and the output is immediately measurable, but speed does not remove the review gate. A well-named set of candidates and consistent native-size route is more reliable than trusting the preview thumbnail.
Retain the original and the approved recipe. If a new layout needs different dimensions, regenerate and inspect again rather than stretching or recompressing the published derivative. Quality control is not a one-time declaration about a source image; it applies to each distinct delivery file.