Why sequence still beats spinning every control at once
Noise reduction averages neighbors. Sharpening amplifies differences. If you sharpen first, the kernel sees speckles as detail and turns them into jewelry.
We run quieting before contrast because a wider curve makes noise more visible afterward. Local punch last so micro contrast rides on the cleaned signal. That order is a opinionated default, not physics law, yet teams keep rediscovering the same stack on timelines.
Tuesday desk example: a 2400 px product still from a phone under mixed LED. Shadows held pepper. Pushing edge emphasis to forty before touching grain quieting produced chrome halos on label type. Dropping edge emphasis to zero, lifting grain quieting to twenty five first, then edge emphasis near twenty five, kept lettering calm.
What each pass is doing in plain language
Grain quieting is a three-by-three box blend on red, green, blue. Strength scales how much each pixel leans toward the local average. Aggressive values smear fine fabric weave.
Global stretch uses a classic curve around mid gray. Positive numbers push highlights and shadows apart; negative numbers flatten the scene like fog rolling in.
Edge emphasis is a small Laplacian style kernel blended with the untouched sample so you dial intensity without swapping tools.
Local punch compares each channel to a tight neighborhood average, then pushes away from that average. Skin pores react fast; use small amounts for portraits.
Quiet failure modes on phone GPUs
Canvas caps vary. A twelve megapixel file might decode, then stutter when four passes touch every pixel each drag event. If the tab hangs, resize offline, re-export near two thousand pixels on the long edge, reload.
CMYK JPEGs or odd ICC tags sometimes decode to black in Safari. Convert to sRGB RGB in desktop software first.
Reading the preview without fooling yourself
Zoom the browser page, not only the canvas card. Mobile pinch changes CSS pixels while the math still runs on the downsampled bitmap you loaded. If the export looks softer than the live view, compare both at native pixel zoom after download.
Band your tests when time allows: save three PNGs at low, medium, and high edge emphasis with identical grain quieting, then flip between them in a desktop viewer. Side-by-side beats trusting memory after the third slider pass.
Displays disagree. A budget IPS panel shows halos before a calibrated mini LED does. Ship the file to the platform you care about, glance on that screen, then decide whether to walk sliders back.
Reality check beside single-purpose pages
Need only blur? Blur keeps the intent honest. Chasing a film look after structural fixes? Stack a graded preset from professional filters once neutrals behave.
| Expectation | Verdict |
|---|---|
| Recover blown sky detail | No. Missing sensor data stays missing. Lower exposure at capture or bracket. |
| Upscale to print poster size | No true super resolution here. Output matches preview resolution. |
| Remove JPEG blockiness entirely | Partial. Quiet specks helps; heavy artifacts need dedicated repair models. |
| Keep EXIF after export | PNG from canvas strips most camera metadata. Use metadata viewer on the source file first if you need serial numbers or lens notes. |
Numbers worth stealing for a first pass
Auto tune lands near thirty edge emphasis, twenty grain quieting, ten global stretch, twenty five local punch. On high ISO night shots, try twenty edge emphasis and thirty five grain quieting before touching local punch.
Flat product flats on gray paper often want five to fifteen global stretch with single-digit local punch so seams stay invisible.
