About imagegrain

A basic noise filter chooses a new random value for every pixel. imagegrain instead builds a correlated texture with visible clusters rather than isolated pixel speckles. The default organic model blurs seeded random noise into round, irregular clumps with no underlying grid. The original classic model places random samples on a grid and smoothly interpolates between them; it remains available, and older preset links keep using it so they reproduce exactly. Fine, medium, and coarse modes change the scale of the pattern in both models.

The goal is a useful digital approximation, not a claim to reproduce a named film stock exactly. Actual film grain depends on emulsion, exposure, development, scanning, and enlargement. Kodak describes film as layers containing random patterns of silver-halide crystals and other chemistry; those physical processes are more complex than a browser filter.

The comparison shown on the homepage uses the built-in sample image at 1600×1000 pixels with fine grain at 40%. The same rendered canvas is exported once as PNG and at 85% quality as JPEG and WebP. In the current test, the files measure 2.45 MB, 0.26 MB, and 0.19 MB respectively.

Those figures demonstrate one controlled example, not a universal saving. Export size changes with image detail, dimensions, grain settings, browser encoder, format, and quality. We state the inputs so the result can be reproduced with the sample-image button in the editor.

The repository includes a DOM-free TypeScript grain package and the source used by the browser editor. The public code makes the deterministic settings, tonal response, and stated limitations inspectable instead of asking readers to trust a black box.

The MCP package remains pre-release until its clean-install and end-to-end client tests are complete; the documentation states that status explicitly.