Dither image

Floyd-Steinberg error diffusion, applied to each of the red, green and blue channels separately. Every pixel is snapped to the nearest allowed level and the rounding error is pushed to its neighbours in the classic 7, 3, 5, 1 sixteenths pattern, which is what produces the grainy, ordered look instead of flat banding.

Levels per colour channel, default 2, sets how many values each channel takes. Two means each channel is fully off or fully on, which gives eight possible colours in total. Four levels per channel gives 64 colours, and so on. The range is 2 to 32, and anything else stops with "levels must be a whole number between 2 and 32".

Not necessarily, and often not. The palette reduction happens in the pixels, but a full-colour format still stores every pixel at full depth, and the scattered dots defeat both JPEG and the row filters PNG leans on. One 343,153 byte PNG came back at 348,090 bytes holding eight colours, larger than it went in. GIF is the format that reflects the smaller palette: the same picture lands at 103,576 bytes, so convert the result there when size is the point.

Yes, wherever the output format holds transparency. Only the colour channels are quantised; the alpha of each pixel is carried across untouched, so a transparent PNG background stays transparent. A GIF is the exception: this encoder writes no transparent index, so a GIF comes back fully opaque. The output uses the input format, named with a -dither suffix, such as art-dither.png, with a lossy target encoded at the Quality option, default 80.

Yes, all of it. Decoding, the error diffusion pass and encoding happen in a Web Worker in this tab. Nothing is uploaded and nothing is kept. Every format Convert image reads is accepted, matched on the file extension, and anything else is refused with a message naming your file and the formats that are readable.