WebP lossy, lossless, near-lossless, and effort tuning

WebP combines VP8 intra-frame lossy compression and LZ77 + entropy lossless compression in one container, supporting an 8-bit alpha channel and multi-frame animation in both modes.

1938×1292 Gate of the Orient night photograph used for WebP encoding — **Input: JPEG**1938 × 1292, 610,191 bytes

1938×1292 WebP output encoded at Quality 82 — **Lossy WebP (Q=82): WebP**1938 × 1292, 402,162 bytes (34.1% smaller than the input JPEG, with source ICC)

Comparing the three WebP encoding modes

Beyond standard Quality and Lossless, libvips exposes near-lossless mode (NearLossless), which is particularly effective on UI screenshots, charts, and icons:

Lossy mode (Lossless = false): Quality controls VP8 quantization (default 75; 78–85 is typical for web photography).
Pure lossless mode (Lossless = true): preserves exact pixel values; Quality controls compression search effort rather than visual loss.
Near-lossless mode (NearLossless = true): uses the lossless VP8L encoder after gently quantizing pixel deltas to improve LZ77 repeatability (Quality sets the preprocessing precision, such as 60 or 80), frequently cutting UI screenshot sizes by 30%–50% with negligible visual change.

Setting	govips `WebpExportParams`	sharp `.webp(options)`	pyvips / vips CLI `webpsave`	Notes
Quality / preprocessing	`Quality` (`0..100`)	`quality` (default 80)	`Q=82` / `--Q 82`	Controls quantization in lossy mode or delta rounding in near-lossless mode
Pure lossless	`Lossless: true`	`lossless: true`	`lossless=True` / `--lossless`	Enables exact VP8L lossless compression
Near-lossless	`NearLossless: true`	`nearLossless: true`	`near_lossless=True` / `--near-lossless`	Pre-smooths pixels before lossless encoding; ideal for UI graphics
CPU effort	`ReductionEffort` (`0..6`, default 4)	`effort` (`0..6`, default 4)	`effort=4` / `--effort 4`	`0` is fastest; `4` is the production default; `6` is slowest
Alpha quality	— (does not expose `alpha_q`; default 100)	`alphaQuality` (`0..100`)	`alpha_q=90` / `--alpha-q 90`	Setting `85–90` on lossy RGBA WebP shrinks translucent masks significantly
Smart chroma subsampling	— (uses libwebp default or `Preset`)	`smartSubsample: true`	`smart_subsample=True`	Higher-quality YUV 4:2:0 chroma downsampling at extra CPU cost
Metadata retention	`Keep` (`ForeignKeep*`)	`.keepIccProfile()` etc.	`keep="icc"` / `--keep icc`	Defaults differ by binding; govips v2.20 combines ICC, EXIF, XMP, IPTC, and other flags
Target byte size	`TargetSize` (bytes)	—	`target_size` / `--target-size`	Version-sensitive: requires `libvips >= 8.17.4`

1. Lossy photography vs near-lossless UI graphics

// Case A: Lossy WebP for photographs (govips v2.20.0 does not expose alpha_q; alpha uses default 100)
lossy := vips.NewWebpExportParams()
lossy.Quality = 82
lossy.ReductionEffort = 4
lossy.StripMetadata = true
data, _, err := image.ExportWebp(lossy)

// Case B: Near-lossless WebP for UI screenshots, diagrams, and badges
nearLossless := vips.NewWebpExportParams()
nearLossless.NearLossless = true
nearLossless.Quality = 60 // 60 applies moderate pixel preprocessing for higher lossless compression
nearLossless.ReductionEffort = 4
nlData, _, err := image.ExportWebp(nearLossless)

import sharp from 'sharp';

// Case A: Lossy WebP with smartSubsample for sharper coloured edges
await sharp('input.jpg')
  .webp({
    quality: 82,
    alphaQuality: 90,
    effort: 4,
    smartSubsample: true,
  })
  .toFile('photo-lossy.webp');

// Case B: Near-lossless WebP for UI diagrams or icons
await sharp('ui-diagram.png')
  .webp({
    nearLossless: true,
    quality: 60,
    effort: 4,
  })
  .toFile('ui-near-lossless.webp');

import pyvips

image = pyvips.Image.new_from_file("input.jpg")

# Case A: Lossy WebP with smart chroma subsampling
image.write_to_file(
    "photo-lossy.webp",
    Q=82,
    alpha_q=90,
    effort=4,
    smart_subsample=True,
    keep="icc",
)

# Case B: Near-lossless WebP
image.write_to_file(
    "ui-near-lossless.webp",
    near_lossless=True,
    Q=60,
    effort=4,
)

# Case A: Lossy WebP
vips webpsave input.jpg photo-lossy.webp --Q 82 --alpha-q 90 --effort 4 --smart-subsample

# Case B: Near-lossless WebP
vips webpsave input.jpg ui-near-lossless.webp --near-lossless --Q 60 --effort 4

2. govips v2.20.0 `Keep` and WebP ICC

WebpExportParams.Keep is a metadata bitmask. Zero means unset, so StripMetadata still decides whether metadata is stripped; an explicit Keep takes precedence:

params := vips.NewWebpExportParams()
params.StripMetadata = true
params.Keep = vips.ForeignKeepIcc
data, _, err := image.ExportWebp(params)

On the current v2.20.0/libvips 8.17.3 baseline, default WebP output retains the source ICC. ForeignKeepNone removes all metadata, while ForeignKeepIcc preserves ICC even with strip enabled. ForeignKeepGainmap requires libvips 8.18, so this baseline does not retain gain maps. Treat sharp, pyvips, and CLI switches and defaults independently.

3. `TargetSize` version boundary (`libvips 8.17.4+`)

govips v2.19.0 adds TargetSize int to WebpExportParams (mapping to target_size in vips_webpsave) so libwebp can run multi-pass encoding to approach a target byte budget.

WebP lossy, lossless, near-lossless, and effort tuning

Comparing the three WebP encoding modes

1. Lossy photography vs near-lossless UI graphics

2. govips v2.20.0 Keep and WebP ICC

3. TargetSize version boundary (libvips 8.17.4+)

Sources

2. govips v2.20.0 `Keep` and WebP ICC

3. `TargetSize` version boundary (`libvips 8.17.4+`)