EXIF auto-orient, orthogonal rotation, and arbitrary angles

In libvips, “rotating an image” maps to three distinct underlying operations depending on the angle and metadata. Mixing them up wastes CPU, risks rotating upright photos twice, or introduces unexpected dark corner triangles.

Input photograph: JPEG1938 × 1292

Arbitrary-angle rotation (30°): WebP2324 × 2088, background #1d2d35 (govips reference output)

Choosing the right rotation operation

Scenario	libvips operation	Interpolated?	Output canvas dimensions	Typical use case
Normalize EXIF Orientation	vips_autorot	No (exact pixel rearrangement)	Swaps width/height on 90°/270°	Normalizing smartphone camera uploads
90° / 180° / 270° orthogonal turn	vips_rot	No (exact pixel rearrangement)	Swaps width/height on 90°/270°	User-triggered “Rotate 90° left/right” actions
Arbitrary angle & scale	vips_similarity / vips_affine	Yes (bilinear/bicubic)	Bounding box expands; corners filled with background	Deskewing, creative layouts, 2×2 affine matrices

1. EXIF Orientation normalization (AutoRotate / autoOrient)

Smartphone cameras typically record sensor pixels in a fixed orientation and store the physical camera tilt in the EXIF Orientation tag (1..8). If you crop or watermark an image before normalizing its orientation, the crop or watermark lands in the wrong corner.

Calling AutoRotate (.autoOrient() or zero-argument .rotate() in sharp) rotates/flips the pixels according to the EXIF tag and resets or removes the EXIF Orientation tag so browsers do not rotate the image a second time.

govips

package main

import (
    "fmt"
    "os"

    "github.com/davidbyttow/govips/v2/vips"
)

func main() {
    if err := vips.Startup(nil); err != nil {
        fmt.Fprintln(os.Stderr, err)
        os.Exit(1)
    }
    defer vips.Shutdown()

    image, err := vips.NewImageFromFile("input.jpg")
    if err != nil {
        fmt.Fprintln(os.Stderr, err)
        os.Exit(1)
    }
    defer image.Close()

    // Rotate/flip pixels according to EXIF Orientation and reset the tag
    if err := image.AutoRotate(); err != nil {
        fmt.Fprintln(os.Stderr, err)
        os.Exit(1)
    }

    data, _, err := image.ExportWebp(vips.NewWebpExportParams())
    if err != nil {
        fmt.Fprintln(os.Stderr, err)
        os.Exit(1)
    }
    if err := os.WriteFile("autorotated.webp", data, 0o644); err != nil {
        fmt.Fprintln(os.Stderr, err)
        os.Exit(1)
    }
}

sharp

import sharp from 'sharp';

// autoOrient() (or zero-arg rotate()) normalizes pixels from EXIF Orientation and resets the tag
await sharp('input.jpg')
  .autoOrient()
  .webp({ quality: 82 })
  .toFile('autorotated.webp');

pyvips

import pyvips

image = pyvips.Image.new_from_file("input.jpg")
upright = image.autorot()
upright.write_to_file("autorotated.webp", Q=82)

vips CLI

#!/usr/bin/env bash
set -euo pipefail

vips autorot input.jpg 'autorotated.webp[Q=82]'

Auto-rotation and AccessSequential

When EXIF Orientation is 1 (normal) or 2 (horizontal mirror), evaluation can stay sequential. Orientations 3..8 require 90°/180°/270° rotation or vertical flips, which read scanlines out of order. Do not call AutoRotate directly on an image opened with AccessSequential when handling arbitrary user uploads — use default random-access loading, or let NewThumbnailWithSizeFromFile / TranscodeStream manage materialization automatically.

2. Orthogonal 90° / 180° / 270° rotation (Rotate / rot)

When rotating by an exact multiple of 90°, use orthogonal rotation. It rearranges pixel coordinates directly without interpolation blur or corner background fill:

govips

// Accepts vips.Angle0, vips.Angle90, vips.Angle180, vips.Angle270 (clockwise)
if err := image.Rotate(vips.Angle90); err != nil {
    return err
}

sharp

// Passing 90, 180, or 270 automatically selects fast orthogonal rotation
await sharp('input.jpg')
  .rotate(90)
  .webp({ quality: 82 })
  .toFile('rot90.webp');

pyvips

import pyvips

image = pyvips.Image.new_from_file("input.jpg")
# Accepts "d0", "d90", "d180", "d270" or helpers rot90() / rot180() / rot270()
rotated = image.rot("d90")
rotated.write_to_file("rot90.webp", Q=82)

vips CLI

vips rot input.jpg 'rot90.webp[Q=82]' d90

3. Arbitrary-angle rotation and 2×2 affine transforms (Similarity / Affine)

For non-orthogonal angles (such as the 30° clockwise rotation shown at the top of this page), libvips uses vips_similarity to compute an expanded bounding box and interpolate pixels. Rotating the 1938 × 1292 source by 30° expands the canvas to 2324 × 2088 and fills the four exposed corner triangles with background (defaulting to black [0, 0, 0] if omitted).

When you also need shear or non-uniform 2×2 matrix transforms ([a, b, c, d]), sharp, pyvips, and vips CLI expose affine. govips v2.19.0 exposes Similarity (and 2D coordinate LUT resampling via Mapim) rather than a public Affine method:

govips

// Similarity(scale, angle, backgroundColor, idx, idy, odx, ody)
bg := &vips.ColorRGBA{R: 29, G: 45, B: 53, A: 255}
if err := image.Similarity(1.0, 30.0, bg, 0, 0, 0, 0); err != nil {
    return err
}

sharp

// 1. Arbitrary-angle rotation with explicit background colour
await sharp('input.jpg')
  .rotate(30, { background: { r: 29, g: 45, b: 53 } })
  .webp({ quality: 82 })
  .toFile('rotate-30.webp');

// 2. General 2×2 affine matrix transform (e.g. horizontal shear)
await sharp('input.jpg')
  .affine([1, 0.15, 0, 1], { background: '#1d2d35' })
  .webp({ quality: 82 })
  .toFile('sheared.webp');

pyvips

import pyvips

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

# 1. Arbitrary-angle rotation
rotated = image.similarity(scale=1.0, angle=30.0, background=[29, 45, 53])
rotated.write_to_file("rotate-30.webp", Q=82)

# 2. General 2×2 affine matrix [a, b, c, d]
sheared = image.affine([1.0, 0.15, 0.0, 1.0], background=[29, 45, 53])
sheared.write_to_file("sheared.webp", Q=82)

vips CLI

# 1. Arbitrary-angle rotation
vips similarity input.jpg 'rotate-30.webp[Q=82]' --angle 30 --background 29,45,53

# 2. General 2×2 affine matrix transform
vips affine input.jpg 'sheared.webp[Q=82]' "1 0.15 0 1" --background 29,45,53

Sources

• libvips autorot documentation
• libvips rot and similarity documentation
• govips v2.19.0 AutoRotate, Rotate, and Similarity
• sharp autoOrient, rotate, and affine