About

JPEG compression relies on 8×8 pixel blocks processed through Discrete Cosine Transform. Databending exploits this structure by corrupting encoded byte sequences between the Start of Scan (SOS, marker 0xFFDA) and End of Image (EOI, marker 0xFFD9). Altering bytes outside this range destroys the file header and renders the image unreadable. This tool operates on decoded pixel data via Canvas getImageData, applying deterministic corruption algorithms: channel displacement shifts R, G, B planes independently, pixel sorting reorders rows by luminance thresholds, and scanline injection inserts random byte blocks into re-encoded JPEG binary. Careless manual hex editing corrupts files irreversibly. The tool preserves structural integrity while maximizing visual distortion.

Results vary based on JPEG quantization tables and image content. High-frequency detail areas (textures, edges) glitch more dramatically than flat color regions. The "Seed" parameter ensures reproducible results. Export at full resolution with no server upload required.

Formulas

Channel shift displaces each color plane independently across the image canvas:

R′(x, y) = R(x + dx_r, y + dy_r)

Pixel sorting compares luminance values calculated from the ITU-R BT.601 standard:

L = 0.299 ⋅ R + 0.587 ⋅ G + 0.114 ⋅ B

Pixels within row are sorted only if their luminance falls in the range L_min ≤ L ≤ L_max. Wave distortion applies a sinusoidal horizontal offset per row:

dx(y) = A ⋅ sin(2π ⋅ yλ+ φ)

Where A = amplitude in pixels, λ = wavelength in pixels, φ = phase offset. The seeded pseudo-random number generator uses a linear congruential generator: x_n+1 = (a ⋅ x_n + c) mod m, with a = 1664525, c = 1013904223, m = 2³², ensuring reproducible results for a given seed value.

Reference Data

Glitch Effect	Technique	Visual Result	Intensity Range	Best For
Channel Shift	Offset RGB planes by dx, dy pixels	Chromatic aberration, color fringing	0 - 200 px	Portraits, high-contrast images
Pixel Sorting	Sort row pixels by luminance L within threshold	Streaked, melted look	0 - 255 threshold	Landscapes, skylines
Scanline Corruption	Inject random bytes into JPEG SOS data	Blocky artifacts, color smearing	1 - 500 corruptions	Any JPEG source
Chunk Displacement	Shift horizontal pixel bands by random offset	Broken, displaced bands	0 - 100 px	Text overlays, geometric images
Wave Distortion	Sine-wave row displacement: dx = A ⋅ sin(2πy / λ)	Wavy, liquid distortion	Amplitude 0 - 80 px	Abstract art, backgrounds
Quantization Smash	Re-encode at quality Q < 10	Heavy DCT block artifacts	Quality 1 - 15	Vaporwave aesthetic
Color Crush	Reduce color depth per channel to n levels	Posterized, banded gradients	2 - 32 levels	Gradients, photographs
Line Injection	Insert solid/noise scanlines at interval k	CRT/VHS scan lines	Every 2 - 20 rows	Retro aesthetic
Bit Shift	Bitwise shift pixel values: v = v << n	Extreme color distortion	1 - 4 bits	Experimental art
Block Shuffle	Randomly swap 8×8 DCT-aligned blocks	Mosaic scrambling	1 - 50% blocks	Censorship effect, chaos
Ghosting	Blend frame with offset copy at α opacity	Motion blur, echo effect	Offset 5 - 50 px	Action photos
Inversion Bands	Invert pixel values in random horizontal strips	Negative color bands	1 - 30 bands	High-contrast images

Frequently Asked Questions

JPEG compression artifacts concentrate at edges and high-frequency detail areas. Images with sharp transitions between colors (text on backgrounds, architectural lines) produce more visible block artifacts when corrupted. Flat, low-contrast images like foggy landscapes resist visible glitching because the DCT coefficients are already near-zero. Increase scanline corruption and channel shift intensity for subtle images.

The tool uses a Linear Congruential Generator (LCG) seeded with your input value. Every random decision (byte injection location, chunk offset magnitude, corruption targets) derives from this deterministic sequence. The same seed + same image + same parameters = identical output every time. Change the seed by even 1 to get a completely different glitch pattern.

No. The tool corrupts pixel data on the Canvas level, not the raw JPEG binary. The final export is a clean re-encode of the corrupted pixel buffer. The output file is always structurally valid. If you select PNG export, no JPEG artifacts appear - only the pixel-level manipulations (channel shift, sorting, wave) are visible.

Canvas API limits vary by browser: Chrome supports approximately 16,384 × 16,384 pixels (268 megapixels), Firefox around 11,180 × 11,180. Images exceeding these limits are automatically downscaled. For best performance, keep images under 4,000 × 4,000 pixels. Processing time scales linearly with pixel count - a 12-megapixel image takes roughly 3× longer than a 4-megapixel image.

Yes. Effects are applied in a fixed pipeline order: Color Crush → Channel Shift → Pixel Sorting → Wave Distortion → Chunk Displacement → Scanline Lines → Block Shuffle. This order matters. Pixel sorting after channel shift produces different results than the reverse. Set any effect's intensity to zero to skip it in the chain.