Starting…

Scheme: catppuccin-mocha

Original Colorized

Drag the handle to compare. Turn on the loupe and touch or hover the preview to inspect pixels.

See the same image snapped to popular palettes

Original plus live-rendered Kanagawa, Catppuccin Mocha, Nord, and Gruvbox examples.

Colorscheme

Loading colorschemes…

Using GPU acceleration

Your browser is using the graphics path, so Image Colorizer can split the image into many small chunks and process them in parallel on your GPU.

Using slower mode

Your browser could not use the faster graphics path, so Image Colorizer is using a slower local fallback. It still works, but large images and big radius values can take longer.

For faster rendering, install the CLI and start the local app:

cargo install image-colorizer
image-colorizer serve

Then open http://localhost:8474.

Why GPU acceleration is faster

The colorizer touches every pixel several times. The GPU version splits that work into thousands of tiny parallel jobs. CPU fallback has far fewer workers available, so big images and large spatial radius values take longer.

What Image Colorizer is doing

Every render runs the image through three stages. The math is the same idea in fast mode and slower mode; the difference is how much work can happen at once.

1

Match each pixel to the palette

Convert source color to Lab, find the closest palette color, apply blend and dither.

CPU fallback

Works through pixels in a narrow stream.

GPU fast path

Many pixels match against the palette at the same time.

2

Smooth color decisions horizontally

Look left and right within the spatial radius to soften abrupt chroma changes.

CPU fallback

Each output pixel samples a row window before moving on.

GPU fast path

Many row windows are averaged in parallel.

3

Smooth vertically and restore luminance

Average the horizontal result up and down, restore the original lightness, and write final pixels.

CPU fallback

Final pixels arrive as the CPU completes each region.

GPU fast path

The final image is produced by lots of small chunks finishing together.

Reference benchmark

Measured with the same colorizer settings on synthetic gradient images. Exact speeds vary by device, but this is the scale of the difference when the browser can use the graphics path.

512 × 512

GPU fast path
15.2 ms
CPU fallback
405.8 ms
About 27× faster

1024 × 1024

GPU fast path
22.9 ms
CPU fallback
1,666.4 ms
About 73× faster

The larger image pulls ahead more because the GPU can keep adding parallel chunks, while CPU fallback has to spend much more time walking the same pixels.