Questions about Bandaid

No. Bandaid is a static web page. Decoding, analysis and export all run in your browser, in a background thread on your own machine. The site never receives your images; you could load it, go offline, and keep working.

Not in the machine-learning sense. There's no trained model and nothing invented. It measures the band pattern with signal processing (medians, a Fourier transform and lock-in demodulation) and divides it out. Because it only ever brightens or darkens whole rows by a measured amount, it can't add detail that wasn't there, and it can't change faces or text.

Flicker banding: horizontal or vertical stripes, often tinted magenta or green, caused by LED lighting or LED screens plus a fast or electronic shutter. It doesn't fix gradient banding (stepped skies from compression), moiré, or sensor pattern noise.

Anything your browser can decode: JPEG, PNG and WebP everywhere, AVIF in most browsers, and HEIC in Safari. Raw files aren't supported, so export a JPEG or TIFF first. Ideally run Bandaid before heavy local edits, so the bands are still a simple brightness change.

The detector needs a sharp, repeating rhythm that shows up across a reasonable part of the frame. Very faint bands confined to a small patch of background can fall below that bar. The bar is set high on purpose, so that clean photos are never altered. Turn on Fix even if no bands are detected in the Correction panel; it still uses the best band spacing it found. Then check the result in split view.

• Spacing: distance from one band to the next, in pixels of your original file. It's set by sensor readout speed and the light's flicker rate.
• Depth: how much the darkest bands are lifted, in stops. 1 stop means twice as bright.
• Confidence: how sure the detector is that a steady band rhythm is present. Below 50%, the photo is treated as clean.
• Coverage: the share of the frame's 16 zones that show the rhythm on their own.

Energy in the row-to-row brightness changes, at every possible band spacing (log scale, spacing in pixels along the bottom). A flicker-banded photo has a tall, narrow spike at its band spacing, marked in magenta, with smaller echoes at ½, ⅓ and ¼ of it (dotted). The shaded strip is where the detector's confidence ramps from 0 to 100%.

A view of how much each part of the photo was changed, over a dim greyscale version of it. Bright stripes are where the fix did the most work. It's a quick way to confirm the correction sits on the bands and not on your subject.

Brightest treats the rows that caught the full light pulse as correct and lifts the dark rows to match, which usually gives the most natural result. Average flattens the bands around their mean, so the overall exposure doesn't change. Use it if Brightest makes the frame look too bright, or if highlights start to clip.

The frame is split into vertical strips, and each gets its own band measurement, blended smoothly into its neighbours. More zones follow bands that change across the frame, such as LED wall tiles with their own timing, or a wall that only covers part of the shot. Fewer zones are steadier on noisy images. 12 is a good default.

Rarely needed. JPEG compression and noise can make the measured depth slightly shallower than the real one, so a little extra can clean up a faint leftover band. Watch the split view: too much and the bands come back reversed.

Each photo keeps its own settings. Use for all photos copies the current photo's settings to everything loaded, and saves them as the default for photos you add later.

For JPEG export, yes: the original EXIF block (camera, lens, settings, capture time, copyright) is copied into the new file. Bandaid resets the orientation tag because the pixels are saved already upright. PNG and WebP don't carry it over.

Yes. Analysis runs on a copy up to 3,000 px on its longest side, but the correction is scaled up and applied to every pixel of the original file.

Skip photos without bands in batch is on by default, so the zip only contains photos that needed fixing (or that you forced). Turn it off to export everything.

• Match the flicker. Use your camera's fine shutter steps ("variable shutter", "high-frequency flicker reduction", "synchro scan") and dial in until the bands disappear in live view. You're aiming for an exact multiple of the light's flicker period.
• Slow down if you can: the more flicker cycles each row sees, the shallower the bands.
• Use the mechanical shutter. It sweeps the frame faster than most electronic shutters, which gives fewer, wider bands.
• Turn on anti-flicker for mains-frequency lights (100/120 Hz). It times each shot to the brightest part of the cycle.

Each frame starts at a different point in the light's on/off cycle, so the stripes land in a different place every frame. Try the Video mode in the shutter simulator. Bandaid works on stills only.