Short answer
Lossy image formats convert RGB to one brightness channel plus two colour-difference channels and usually store the colour channels at reduced resolution — chroma subsampling — because vision is less sharp for colour than for brightness. That is invisible in most photographs but smears colour along sharp edges, which shows up as fringes around red text, logos and flat graphics. Lossy WebP always subsamples; JPEG and AVIF can be set not to. Formats also differ in bit depth and in how they tag their colour space, which decides whether wide-gamut and HDR colour survives.
JPEG files in the JFIF convention convert RGB into Y′CbCr — a luma channel and blue-difference and red-difference channels — using the BT.601 weights. The eye resolves fine detail mostly through luminance, so the two chroma channels can be stored at half resolution horizontally (4:2:2) or in both directions (4:2:0) with little visible loss in photographs. The saving is substantial: 4:2:0 stores a quarter as many samples for each chroma channel. Lossy WebP, based on the VP8 video codec, works exclusively in 8-bit Y′CbCr 4:2:0. AVIF inherits AV1's profiles: the Main profile, which the AVIF baseline requires, covers 4:2:0 at 8 or 10 bits, while the High profile adds full-resolution 4:4:4.
Chroma subsampling shares one colour value between neighbouring pixels. In a photograph neighbouring pixels usually have similar colours, so nothing is lost that anyone would notice. At a sharp boundary between two saturated colours, or between saturated colour and white or black, the shared colour sample straddles the edge and bleeds: red text on blue gets a dark fringe, thin coloured lines lose saturation, and small coloured type looks blurry while black type stays crisp. Screenshots, charts, UI mock-ups and logos are the worst cases. For those, a lossless format (PNG or lossless WebP, both RGB) or a 4:4:4 JPEG or AVIF avoids the problem.
A quick diagnostic: zoom to 400% on a red-on-white edge. A soft colour halo one or two pixels wide, while the luminance edge is sharp, is subsampling.
Eight bits per channel is enough for most sRGB images, but gradients in wide-gamut or HDR images need more because each code step spans a larger colour difference. Lossy WebP is fixed at 8 bits; JPEG as normally used is 8-bit; AVIF supports 10 and 12 bits, which is why it is the more common web choice for HDR stills. Heavy compression also quantises colour detail, so skies and skin can show blotches or bands at low quality settings; the chroma channels are usually compressed harder than luma, which can shift subtle hues in smooth areas.
Compression artefacts are only half the story; the other half is metadata. JPEG and WebP can embed an ICC profile; AVIF signals colour through coding-independent code points (primaries, transfer and matrix) and can also carry an ICC profile; PNG's third edition, a W3C Recommendation since June 2025, adds a cICP chunk and mastering-display metadata alongside its long-standing iCCP and sRGB chunks. Many image pipelines — CDNs, CMS resizers, 'optimisers' — strip metadata to save bytes. For an sRGB image that is harmless, since untagged means sRGB. For a Display P3 or HDR image it silently changes the colours. Configure optimisers to keep colour profiles, or convert to sRGB first.
| Format | Chroma subsampling | Bit depth (typical web use) | Colour tagging |
|---|---|---|---|
| JPEG (JFIF) | Optional; 4:2:0 common | 8-bit | ICC profile |
| WebP, lossy | Always 4:2:0 | 8-bit | ICC profile |
| WebP, lossless | None (RGBA) | 8-bit | ICC profile |
| AVIF, baseline profile | 4:2:0 or monochrome | 8 or 10-bit | CICP code points and/or ICC |
| AVIF, advanced profile | Up to 4:4:4 | 8 or 10-bit | CICP code points and/or ICC |
| PNG | None (RGB/RGBA) | 8 or 16-bit | iCCP, sRGB, and cICP (third edition) |
Why: Lossy compression with 4:2:0 chroma subsampling across sharp coloured edges.
Fix: Export graphics as PNG or lossless WebP, or as JPEG/AVIF with 4:4:4 chroma.
Why: The optimiser stripped the embedded profile.
Fix: Enable profile preservation or convert to sRGB before upload.
Why: 8-bit encoding at low quality, often in a wide-gamut space.
Fix: Raise quality, add slight noise, or use a 10-bit AVIF.
Each statement is labelled by kind — established fact, a standard’s requirement, observed market data, a convention, or Colourwise’s own interpretation or analysis — with the strength of the evidence behind it.
FactStrong evidence
Lossy WebP works exclusively with 8-bit Y′CbCr 4:2:0 images, following the VP8 bitstream, while lossless WebP works with RGBA.
Source: WebP frequently asked questions
StandardStrong evidence
AV1's Main profile supports 4:2:0 or monochrome at 8 or 10 bits and the High profile adds 4:4:4; the AVIF baseline profile requires the AV1 Main profile and the advanced profile the High profile.
Source: AV1 Bitstream Specification and AV1 Image File Format (AVIF) v1.2.0
StandardStrong evidence
JFIF-format JPEG files store colour as Y′CbCr defined per CCIR (ITU-R BT.) 601.
Source: JPEG File Interchange Format, version 1.02 (E. Hamilton, C-Cube Microsystems, 1992)
StandardStrong evidence
The third edition of the PNG specification, a W3C Recommendation of 24 June 2025, adds the cICP chunk for colour signalling alongside the iCCP (embedded ICC profile) and sRGB chunks.
Source: Portable Network Graphics (PNG) Specification (Third Edition)
Colourwise interpretationModerate evidence
Chroma subsampling is rarely visible in photographs but commonly visible on sharp saturated edges in graphics, screenshots and text, which are better stored losslessly or at 4:4:4.
Based on: Follows from how 4:2:0 shares one chroma sample among four pixels, as specified for WebP and AV1; not a measured perceptual study.
Reviewed 29 September 2026. Colourwise summarises its sources in its own words and does not reproduce standards text or proprietary colour data. Spotted an error? Tell us.