Databend

Why real databending doesn’t look like noise

The original technique was crude and brilliant: open a JPEG in a text editor, type over some bytes, save, reopen. What comes out is never a sprinkle of wrong pixels. It is blocks — smeared, repeated, sheared, colour-torn.

That is because a compressed file is stateful. The decoder doesn’t store pixels; it stores instructions that only mean anything relative to what came before. Damage one byte and you haven’t corrupted one pixel, you have desynchronised the decoder, and every block after it decodes against the wrong reference until something resynchronises it. The damage propagates forward with the structure of the format, not the structure of the image.

Reproducing that convincingly means corrupting something block-aligned rather than adding noise, which is the distinction between this and the many effects that draw torn rectangles.

What the six modes are named for

Blocks is the classic macroblock failure: a block loses its reference and holds, smears, or collapses to a single flat value across the whole cell. That flat cell is what a decoder produces when only the block’s average survives, and it is the most recognisable datamosh artefact there is.

Bitplane damage is confined to one level of significance. Corrupt the top bit of a byte and the value jumps half its range — catastrophic, posterised, violent. Corrupt the bottom bit and you get something indistinguishable from fine grain. The bitplane you choose is a dial from “barely there” to “unrecognisable” with very little in between.

Bit-shift moves every byte’s bits along, which wraps tone around on itself: bright becomes dark discontinuously rather than gradually, so smooth gradients cut into hard terraces. Header-skew is the one that shears — tell a decoder the wrong row width and every row starts one step further along than it should, so the whole image slants and wraps. Channel-split offsets one colour channel’s data from the others, which is where the colour-tear fringing comes from. Byte-swap exchanges adjacent bytes, scrambling at the finest scale the format has.

Databend versus Compression

These are two different effects and it is worth knowing which you want. Databend corrupts data. Compression runs a real 8×8 DCT — forward transform, quantise against actual JPEG tables, inverse — so its artefacts are ringing, blocking and colour subsampling that appear because the image was genuinely compressed too hard.

If you want the picture to look badly saved, that’s Compression. If you want it to look damaged after it was saved, that’s Databend. Stacking them in that order — compress, then bend — is closer to what actually happens to a file in the wild than either alone.

When you’d reach for it

When you want the failure to look like a machine’s failure rather than a designer’s. Databend’s artefacts have a logic — they align, they propagate, they respect a block grid — and that internal consistency is what makes them read as real rather than decorative.

Every option, from the app

Generated from the registry the app publishes, so this list is whatever the current build actually ships — not a copy of it.

Mode

BlocksBitplaneBit-shiftHeader-skewChannel-splitByte-swap

Related effects

• Pixel SortA true per-span pixel sort along rows or columns, bounded by luma thresholds — sortable by luma, brightness, hue, saturation or a single RGB channel.
• DitherOrdered dithering: Bayer and blue-noise threshold masks quantised to real machine palettes — Game Boy, PICO-8, CGA, EGA, NES, C64, ZX Spectrum and more.
• ASCIIImage to text art: per-cell luminance mapped through fifteen character ramps via a font atlas, with mono, source-colour, duotone and ANSI-16 colour.

Open the app to try it; the free effects work with no signup. Open V3KTR · all the FX · how it works