Read Privacy, Terms, and Disclaimer — especially for cookies, ads, and third-party APIs.

Tool guide

About RGBV3D

RGBV3D is PracticalKit’s strictly lossless image container. This page lets you encode PNG/JPEG inputs to .rgbv3d and decode back with pixel-exact round-trip when the profile allows.

Unlike generic “online compressors”, RGBV3D documents its bitstream: classic predictor + Deflate, or vector-v2 with direction codebook plus exact residuals. We do not promise smaller files than PNG on every image—noise-heavy scans often expand—but UI screenshots and smooth gradients sometimes win.Before you archive production assets, read the format specification (container fields, profile ids) and the benchmark table (four fixture categories including deliberate losses). The encoder on this page shows size versus uncompressed RGB immediately after encode; treat that ratio as a local hint, not a marketing claim.PracticalKit pairs this tool with CLI packages under @rgbv3d/core so the same profiles can be batch-tested outside the browser. Memory and WASM limits still apply in-tab—start under 4K resolution when experimenting. Collaborators can decode your .rgbv3d samples here or via CLI without uploading files to our origin. If you only need resize or WebP, use the image tool first and return here for the lossless container.

Common use cases

  • Experiment with a research codec without installing CLI tools—upload, encode, download .rgbv3d, decode again to verify pixels.
  • Compare encoded size against the same PNG export from our image tool for A/B archive decisions.
  • Share sample .rgbv3d files with collaborators who can use the CLI package @rgbv3d/core or this web decoder.
  • Reproduce CI fixtures: encode a known PNG, hash the RGB buffer after decode, assert equality in your pipeline.
  • Run one noisy scan and one UI screenshot from the benchmark categories so you see ratio differences firsthand.

RGBV3D lossless

Encode raster images to the vector-primary lossless profile in your browser.

Encode and decode run locally in your browser.

Encode to .rgbv3d

8 vs 9 bits = 256 vs 512 direction bins on the sphere. Each 16×16 block still picks the smallest lossless mode (vector, Sub+deflate, plane stack, or raw). Previously, if classic Sub+deflate won for the whole image, 8/9 were ignored and outputs matched — this page fixes that by always using the block-adaptive path.

Browse / decode

Decode a .rgbv3d file and preview the image below — no server upload.

Video containers and benchmarks: build rgbv3d-ecosystem/packages/cli (npm run build).

  1. 1

    Pick a profile

    Classic (profile 0) uses spatial prediction + Deflate—closest to PNG semantics; prefer it for noisy photos. Vector-v2 (profile 1) quantizes direction and norm proxy but stores per-channel residuals so decode is mathematically lossless. Start with classic as a baseline, then switch the same image to vector-v2 and compare size and time.

  2. 2

    Upload and encode

    Choose a PNG or JPEG. The tool rasterizes to RGB24 in-browser, runs encodeArgbLossless, and shows raw size, .rgbv3d size, and ratio vs uncompressed RGB. JPEG inputs are decoded first—any loss already present remains; RGBV3D will not invent detail.

  3. 3

    Inspect the size story

    If the ratio looks worse than PNG, open /rgbv3d/benchmark/ and match your content type to a fixture row. Grain and scan categories explain expansion without blaming the UI.

  4. 4

    Verify lossless decode

    Download .rgbv3d, reload it in decode mode, and visually compare—or export PNG and diff externally with ImageMagick or a short script. Keep originals when testing new profiles.

Tips and limits

  • Grainy photos and ISO-noise frames often compress worse than PNG; that is expected, not a bug.
  • Large images (>4096 px) may hit WASM/RAM limits; try half resolution first.
  • See /rgbv3d/benchmark/ for reproducible sample numbers—not marketing percentages.
  • For sharing outside the RGBV3D toolchain, decode then export PNG from the image tool.
  • Click-pulse on the home lattice is decorative; this page is where encode work happens.
  • Crop first in the image tool if you do not want empty margins packed into the container.

Frequently asked questions

Is RGBV3D lossy?
Profiles 0 and 1 in this repository are strict lossless: decode(encode(pixels)) equals the rasterized input. Near-loss profiles are reserved for future containers and will be labeled explicitly.
Why is my .rgbv3d larger than PNG?
High-frequency noise defeats prediction and direction coding. The benchmark page lists categories where vector-v2 helps vs hurts.
Are files uploaded to PracticalKit servers?
No. Encoding uses Web APIs and @rgbv3d/core in your tab. See Privacy for optional analytics only—kept separate from the tool’s compute path.
Should I replace PNG with RGBV3D everywhere?
No. Use RGBV3D when you need a documented research container and can keep the toolchain. Keep PNG (or AVIF lossless) for interchange with tools that do not speak .rgbv3d.

About this tool page

Files, PDFs, and media in the browser

“RGBV3D” reads files you select in the browser. Where the implementation allows, bytes are processed with Web APIs or WASM on your device without PracticalKit storing the file on our servers for the conversion itself. You trigger result downloads; we do not keep training copies or a server-side queue for this tool’s conversions.

PDF, image, audio, and video workflows can be memory-heavy. Success depends on file size, resolution, encoding, and device RAM. If the browser aborts or freezes, try a smaller asset, a simpler format, or desktop software for very large productions. Encrypted PDFs, complex forms, and hundreds-of-megabyte scanned books are known edge cases that belong on streaming desktop tools.

Policies and editorial pages

Related tools