Usage • Installation • Quick start • Demos • API • Perf • Changelog • FAQ • Contributing

📷 The fastest and most versatile JavaScript EXIF reading library.

Try it yourself - demo page & playground.

Features

Works everywhere, parses anything you throw at it.

• 🏎️ Fastest EXIF lib: +-1ms per file
• 🗃️ Any input: buffers, url, <img> tag, and more
• 📷 Files: .jpg, .tif, .png, .heic, .avif, .iiq
• 🔎 Segments: TIFF (EXIF, GPS, etc...), XMP, ICC, IPTC, JFIF, IHDR
• 📑 Reads only first few bytes
• 🔬 Skips parsing tags you don't need
• ✨ Isomorphic: Browser & Node.js
• 🗜️ No dependencies
• 🖼️ Extracts thumbnail
• 💔 Salvages broken files
• 🧩 Modular
• 📚 Customizable tag dictionaries
• 📦 Bundled as UMD/CJS or ESM
• ✔ Tested and benchmarked
• 🤙 Promises
• 🕸 Supports even ~IE11~ IE10

You don't need to read the whole file to tell if there's EXIF in it. And you don't need to extract all the data when you're looking for just a few tags. Exifr just jumps through the file structure, from pointer to pointer. Instead of reading it byte by byte, from beginning to end.

Exifr does what no other JS lib does. It's efficient and blazing fast!

| Segments | JPEG | TIFF / IIQ | HEIF (HEIC, AVIF) | PNG | |-|-|-|-|-| | EXIF/TIFF, GPS | ✔ | ✔ | ✔ | ✔ | | XMP | ✔ | ✔ | ❌ | ✔ | | IPTC | ✔ | ✔ | ❌ | 🟡 (If it's a part of IHDR) | | ICC | ✔ | ✔ | ✔ | ✔ (Node.js only, requires zlib) | | Thumbnail | ✔ | ❌ | ❌ | ❌ | | JFIF (JPEG header) | ✔ | ⚫ | ⚫ | ⚫ | | IHDR (PNG header) | ⚫ | ⚫ | ⚫ | ✔ |

Usage

file can be any binary format (Buffer, Uint8Array, Blob and more), <img> element, string path or url.

options specify what segments and blocks to parse, filters what tags to pick or skip.

| API | Returns | Description | |-|-|-| |exifr.parse(file)|object|Parses IFD0, EXIF, GPS blocks| |exifr.parse(file, true)|object|Parses everything| |exifr.parse(file, ['Model', 'FNumber', ...])|object|Parses only specified tags| |exifr.parse(file, {options})|object|Custom settings| |exifr.gps(file)|{latitude, longitude}|Parses only GPS coords| |exifr.orientation(file)|number|Parses only orientation| |exifr.rotation(file)|object|Info how to rotate the photo| |exifr.thumbnail(file)|Buffer\|Uint8Array binary|Extracts embedded thumbnail| |exifr.thumbnailUrl(file)|string Object URL|Browser only| |exifr.sidecar(file)|object|Parses sidecar file|

Installation

npm install exifr

Exifr comes in three prebuilt bundles. It's a good idea to start development with full and then scale down to lite, mini, or better yet, build your own around modular core.

// Modern Node.js can import CommonJS
import exifr from 'exifr' // => exifr/dist/full.umd.cjs
// Explicily import ES Module
import exifr from 'exifr/dist/full.esm.mjs' // to use ES Modules
// CommonJS, old Node.js
var exifr = require('exifr') // => exifr/dist/full.umd.cjs

<!-- ES Module in modern browsers -->
<script type="module">import exifr from 'node_modules/exifr/dist/lite.esm.js';</script>
<!-- classic UMD script -->
<script src="https://cdn.jsdelivr.net/npm/exifr/dist/lite.umd.js"></script>
<!-- IE10 & old browsers. You also need Promise polyfill -->
<script src="https://cdn.jsdelivr.net/npm/exifr/dist/lite.legacy.umd.js"></script>

Browsers: lite and mini are recommended because of balance between features and file size. UMD format attaches the library to global window.exifr object.

IE & old browsers: legacy builds come bundled with polyfills. Learn more.

Bundles & formats

• full - Contains everything. Intended for use in Node.js.
• lite - Reads JPEG and HEIC. Parses TIFF/EXIF and XMP.
• mini - Stripped down to basics. Parses most useful TIFF/EXIF from JPEGs. Has no tag dictionaries.

Of course, you can use the full version in browser, or use any other build in Node.js.

• ESM - Modern syntax for use in modern browsers and Node.js. Uses import syntax.
• UMD - Universal format for browsers and Node.js. Supports CJS require('exifr'), AMD/RequireJS and global window.exifr.
• legacy UMD - For use in older browsers (up to IE10). Bundled with polyfills & shims, except for Promise polyfill. Learn more here.

| | full | lite | mini | core | |-----------------|------|------|------|------| | chunkedfile readers | BlobReaderUrlFetcher (+ Node.js)FsReaderBase64Reader | BlobReaderUrlFetcher (Browser only) | BlobReader | none | | file parsers | *.jpg*.heic*.tif/*.iiq*.png | *.jpg*.heic | *.jpg | none | | segmentparsers | TIFF (EXIF)IPTCXMPICCJFIFIHDR | TIFF (EXIF)XMP | TIFF (EXIF) | none | | dictionaries | TIFF (+ less frequent tags)IPTCICCJFIFIHDR | only TIFF keys(IFD0, EXIF, GPS) | none | none | | size +- | 73 Kb | 45 Kb | 29 Kb | 15 Kb | | gzipped | 22 Kb | 12 Kb | 8 Kb | 4 Kb | | file | full.esm.jsfull.esm.mjsfull.umd.jsfull.umd.cjsfull.legacy.umd.js | lite.esm.jslite.esm.mjslite.umd.jslite.umd.cjslite.legacy.umd.js | mini.esm.jsmini.esm.mjsmini.umd.jsmini.umd.cjsmini.legacy.umd.js | Learn more |

ESM, .js .mjs .cjs extensions, "main", "module", "type":"module"

TL;DR: All bundles are available in two identical copies. .mjs and .js for ESM. .cjs and .js for UMD. Pick one that works with your tooling or webserver.

Current state of ESM is complicated. Node.js can already handle ESM files with .mjs extension and modules with "type":"module" in package.json. Turns out the "type":"module" approach alone is not yet ready for production. Some bundlers and tools may work or break with .mjs extension, whereas it's important for Node.js. The same applies to the new .cjs extension (introduced in Node.js 13).

The library is written in ESM, with .mjs extensions and transpiled to both ESM and UMD formats.

The "main" (field in package.json) entry point is now full.umd.cjs but you can still use ESM by explicitly importing full.esm.mjs. "module" field (used by some tools) points to full.esm.mjs.

If your webserver isn't configured to handle .mjs or .cjs files you can use their identical .js clone. For example full.esm.mjs is identical to full.esm.js. So is lite.esm.cjs to lite.esm.js. Just pick one that fits your tools or environment.

Named exports vs default export

Exifr exports both named exports and a default export - object containing all the named exports.

You can use import * as exifr from 'exifr' as well as import exifr from 'exifr' (recommended).

Examples

// exifr reads the file from disk, only a few hundred bytes.
exifr.parse('./myimage.jpg')
  .then(output => console.log('Camera:', output.Make, output.Model))
// Or read the file on your own and feed the buffer into exifr.
fs.readFile('./myimage.jpg')
  .then(exifr.parse)
  .then(output => console.log('Camera:', output.Make, output.Model))

Extract only certain tags

// only GPS
let {latitude, longitude} = await exifr.gps('./myimage.jpg')
// only orientation
let num = await exifr.orientation(blob)
// only three tags
let output = await exifr.parse(file, ['ISO', 'Orientation', 'LensModel'])
// only XMP segment (and disabled TIFF which is enabled by default)
let output = await exifr.parse(file, {tiff: false, xmp: true})

Extracting thumbnail

let thumbBuffer = await exifr.thumbnail(file)
// or get object URL (browser only)
img.src = await exifr.thumbnailUrl(file)

Web Worker

let worker = new Worker('./worker.js')
worker.postMessage('../test/IMG_20180725_163423.jpg')
worker.onmessage = e => console.log(e.data)
// tip: try Transferable Objects with ArrayBuffer
worker.postMessage(arrayBuffer, [arrayBuffer])

// worker.js
importScripts('./node_modules/exifr/dist/lite.umd.js')
self.onmessage = async e => postMessage(await exifr.parse(e.data))

UMD in Browser

<img src="./myimage.jpg">
<script src="./node_modules/exifr/dist/lite.umd.js"></script>
<script>
  let img = document.querySelector('img')
  window.exifr.parse(img).then(exif => console.log('Exposure:', exif.ExposureTime))
</script>

ESM in Browser

<input id="filepicker" type="file" multiple>
<script type="module">
  import exifr from './node_modules/exifr/dist/lite.esm.js'
  document.querySelector('#filepicker').addEventListener('change', async e => {
    let files = Array.from(e.target.files)
    let exifs = await Promise.all(files.map(exifr.parse))
    let dates = exifs.map(exif => exif.DateTimeOriginal.toGMTString())
    console.log(`${files.length} photos taken on:`, dates)
  })
</script>

Demos

• playground
• examples/thumbnail.html, code Extracts and displays embedded thumbnail.
• examples/orientation.html, code Extracts orientation and rotates the image with canvas or css.
• examples/depth-map-extraction.html, code Extracts and displays depth map.
• benchmark/gps-dnd.html, code Drag-n-Drop multiple photos and mesure the time and RAM it took to extract GPS. Then they're marked on a map.
• examples/worker.html, code Parsing file in WebWorker.
• examples/legacy.html, code Visit in IE10/IE11,
• benchmark/formats-reading.html, code Compares reading speed of various input types.

and a lot more in the examples/ folder

API

parse(file[, options])

Returns: Promise<object | undefined>

Accepts file (in any format), parses it and returns exif object. Optional options argument can be specified.

gps(file)

Returns: Promise<object | undefined>

Only extracts GPS coordinates.

Uses pick/skip filters and perf improvements to only extract latitude and longitude tags from GPS block. And to get GPS-IFD pointer it only scans through IFD0 without reading any other unrelated data.

Check out examples/gps.js to learn more.

orientation(file)

Returns: Promise<number | undefined>

Only extracts photo's orientation.

rotation(file)

Returns: Promise<object | undefined>

Only extracts photo's orientation. Returns object with instructions how to rotate the image:

• deg <number>: angle in degrees (i.e. 180), useful for css transform: rotate()
• rad <number>: angle in radians (i.e. 3.141592653589793) useful for canvas' ctx.rotate()
• scaleX <number>: image is (-1) or isn't (1) mirrored horizontally
• scaleY <number>: image is (-1) or isn't (1) mirrored upside down
• dimensionSwapped <boolean>: image is rotated by 90° or 270°. Fixing rotation would swap width and height.
• css <boolean>: can/can't be rotated with CSS and transform: rotate() (important for ios Safari)
• canvas <boolean>: can/can't be rotated with canvas and ctx.rotate() (important for ios Safari)

Warning: Some modern browsers autorotate <img> elements, background-image and/or data passed to <canvas> without altering the EXIF. The behavior is extra quirky on iOs 13.4 Safari and newer (though not on macos). You may end up with over-rotated image if you don't handle this quirk. See examples/orientation.html to learn more.

let r = await exifr.rotation(image)
if (r.css) {
  img.style.transform = `rotate(${r.deg}deg) scale(${r.scaleX}, ${r.scaleY})`
}

thumbnail(file)

Returns: Promise<Buffer | Uint8Array | undefined>

Extracts embedded thumbnail from the photo, returns Uint8Array.

Only parses as little EXIF as necessary to find offset of the thumbnail.

Check out examples/thumbnail.html and examples/thumbnail.js to learn more.

thumbnailUrl(file)

Returns: Promise<string | undefined> browser only

Exports the thumbnail wrapped in Object URL. The URL has to be revoked when not needed anymore.

sidecar(file[, options[, type]])

Returns: Promise<object | undefined> full bundle only

Parses sidecar file, i.e., an external metadata usually accompanied by the image file. Most notably .xmp or .icc.

Third argument is optional but advised if you know the segment type you're dealing with and want to improve performance. Otherwise exifr tries to infer the type from file extension (if file is path or url) and/or randomly tries all parsers at its disposal.

exifr.sidecar('./img_1234.icc')
exifr.sidecar('./img_1234.icc', {translateKeys: false})
exifr.sidecar('./img_1234.colorprofile', {translateKeys: false}, 'icc')

Exifr class

Aforementioned functions are wrappers that internally:

1. instantiate new Exifr(options) class
2. call .read(file) to load the file
3. call .parse() or .extractThumbnail() to get an output

You can instantiate Exif yourself to parse metadata and extract thumbnail efficiently at the same time. In Node.js it's also necessary to close the file with .file.close() if it's read in the chunked mode.

let exr = new Exifr(options)
await exr.read(file)
let output = await exr.parse()
let buffer = await exr.extractThumbnail()
await exr.file?.close?.()

file argument

• string
  • file path
  • URL, Object URL
  • Base64 or Base64 URL
• Buffer
• ArrayBuffer
• Uint8Array
• DataView
• Blob, File
• <img> element

options argument

• array of tags to parse, shortcut for options.pick
• true shortcut to parse all segments and blocks
• object with granular settings

All other and undefined properties are inherited from defaults:

let defaultOptions = {
  // Segments (JPEG APP Segment, PNG Chunks, HEIC Boxes, etc...)
  tiff: true,
  xmp: false,
  icc: false,
  iptc: false,
  jfif: false, // (jpeg only)
  ihdr: false, // (png only)
  // Sub-blocks inside TIFF segment
  ifd0: true, // aka image
  ifd1: false, // aka thumbnail
  exif: true,
  gps: true,
  interop: false,
  // Other TIFF tags
  makerNote: false,
  userComment: false,
  // Filters
  skip: [],
  pick: [],
  // Formatters
  translateKeys: true,
  translateValues: true,
  reviveValues: true,
  sanitize: true,
  mergeOutput: true,
  silentErrors: true,
  // Chunked reader
  chunked: true,
  firstChunkSize: undefined,
  firstChunkSizeNode: 512,
  firstChunkSizeBrowser: 65536, // 64kb
  chunkSize: 65536, // 64kb
  chunkLimit: 5,
  httpHeaders: {},
}

Tag filters

Exifr can avoid reading certain tags, instead of reading but not including them in the output, like other exif libs do. For example MakerNote tag from EXIF block is isually very large - tens of KBs. Reading such tag is a waste of time if you don't need it.

Tip: Using numeric tag codes is even faster than string names because exifr doesn't have to look up the strings in dictionaries.

options.pick

Type: Array<string|number>

Array of the only tags that will be parsed.

Specified tags are looked up in a dictionary. Their respective blocks are enabled for parsing, all other blocks are disabled. Parsing ends as soon as all requested tags are extracted.

// Only extracts three tags from EXIF block. IFD0, GPS and other blocks disabled.
{pick: ['ExposureTime', 'FNumber', 'ISO']}
// Only extracts three tags from EXIF block and one tag from GPS block.
{pick: ['ExposureTime', 'FNumber', 'ISO', 'GPSLatitude']}
// Extracts two tags from GPS block and all of IFD0 and EXIF blocks which are enabled by default.
{gps: {pick: ['GPSLatitude', 0x0004]}}

options.skip

Type: Array<string|number> Default: ['MakerNote', 'UserComments']

Array of the tags that will not be parsed.

By default, MakerNote and UserComment tags are skipped. But that is configured elsewhere.

// Skips reading these three tags in any block
{skip: ['ImageWidth', 'Model', 'FNumber', 'GPSLatitude']}
// Skips reading three tags in EXIF block
{exif: {skip: ['ImageUniqueID', 42033, 'SubSecTimeDigitized']}}

Segments & Blocks

EXIF became synonymous for all image metadata, but it's actually just one of many blocks inside TIFF segment. And there are more segment than just TIFF.

Segments (JPEG APP Segments, HEIC Boxes, PNG Chunks)

Jpeg stores various formats of data in APP-Segments. Heic and Tiff file formats use different structures or naming conventions but the idea is the same, so we refer to TIFF, XMP, IPTC, ICC and JFIF as Segments.

• options.tiff type bool|object|Array default: true TIFF APP1 Segment - Basic TIFF/EXIF tags, consists of blocks: IFD0 (image), IFD1 (thumbnail), EXIF, GPS, Interop
• options.jfif type bool default: false JFIF APP0 Segment - Additional info
• options.xmp type bool default: false XMP APP1 Segment - additional XML data
• options.iptc type bool default: false IPTC APP13 Segment - Captions and copyrights
• options.icc type bool default: false ICC APP2 Segment - Color profile
• options.ihdr type bool default: true (only for PNG) PNG Header chunk - Basic file info

TIFF IFD Blocks

TIFF Segment consists of various IFD's (Image File Directories) aka blocks.

• options.ifd0 (alias options.image) type bool|object|Array default: true IFD0 - Basic info about the image
• options.ifd1 (alias options.thumbnail) type bool|object|Array default: false IFD1 - Info about embedded thumbnail
• options.exif type bool|object|Array default: true EXIF SubIFD - Detailed info about photo
• options.gps type bool|object|Array default: true GPS SubIFD - GPS coordinates
• options.interop type bool|object|Array default: false Interop SubIFD - Interoperability info

Notable TIFF tags

Notable large tags from EXIF block that are not parsed by default but can be enabed if needed.

• options.makerNote type: bool default: false 0x927C MakerNote tag
• options.userComment type: bool default: false 0x9286 UserComment tag

XMP

Extracted XMP tags are grouped by namespace. Each ns is separate object in output. E.g. output.xmlns, output.GPano, output.crs, etc...

For XMP Extended see options.multiSegment

Exifr contains minimalistic opinionated XML parser for parsing data from XMP. It may not be 100% spec-compliant, because XMP is based on XML which cannot be translated 1:1 to JSON. The output is opinionated and may alter or simplify the data structure. If the XMP parser doesn't suit you, it can be disabled by setting options.xmp.parse to false. Then a raw XMP string will be available at output.xmp.

Caveats & XML to JSON mapping

1. Tags with both attributes and children-value are combined into object.
2. Arrays (RDF Containers) with single item are unwrapped. The single item is used in place of the array.
3. If options.mergeOutput:false: Tags of tiff namespace (<tiff:Model>) are merged into output.ifd0. Likewise exif ns is merged into output.exif.

<rdf:Description foo:name="Exifr">
  <foo:author>Mike Kovařík</foo:author>
  <foo:description xml:lang="en-us">Some string here</foo:description>
  <foo:formats><rdf:Seq><rdf:li>jpeg</rdf:li></rdf:Seq></foo:formats>
  <foo:segments><rdf:Seq><rdf:li>xmp</rdf:li><rdf:li>tiff</rdf:li><rdf:li>iptc</rdf:li></rdf:Seq></foo:segments>
</rdf:Description>

parses as:

{
  name: 'Exifr', // attribute belonging to the same namespace
  author: 'Mike Kovařík', // simple tag of the namespace
  description: {lang: 'en-us', value: 'Some string here'}, // tag with attrs and value becomes object
  formats: 'jpeg', // single item array is unwrapped
  segments: ['xmp', 'tiff', 'iptc'] // array as usual
}

options.multiSegment

Type: bool Default: false

Enables looking for more than just a single segment of ICC or XMP (XMP Extended).

In some rare cases the photo can contain additional layers, embedded images, or metadata that doesn't fit inside single 64kB (JPEG) segment.

Side effect: Disables chunked reading. The whole file has to be read to locate all segments.

When is it useful:

• VR photos with combination of left/right eye (XMP Extended)
• "Portrait mode" photo that contains depth map (XMP Extended)
• Photos with custom ICC color profile

Sub-options:

• options.xmp.multiSegment
• options.icc.multiSegment

Shortcuts

options.tiff serves as a shortcut for configuring all TIFF blocks:

• options.tiff = true enables all TIFF blocks (sets them to true).
• options.tiff = false disables all TIFF blocks (sets them to false) except for those explicitly set to true in options.
• options.tiff = {...} applies the same sub-options to all TIFF blocks that are enabled.

options.tiff = false can be paired with any other block(s) to disable all other blocks except for said block.

{interop: true, tiff: false}
// is a shortcut for
{interop: true, ifd0: false, exif: false, gps: false, ifd1: true}

Each TIFF block and the whole tiff segment can also be configured with object or array, much like the options argument.

• object - enabled with custom options - filters (pick, skip) and formatters (translateKeys, translateValues, reviveValues)
• array - enabled, but only picks tags from this array

TIFF blocks automatically inherit from options.tiff and then from options.

// Only extract FNumber + ISO tags from EXIF and GPSLatitude + GPSLongitude from GPS
{
  exif: true, gps: true,
  pick: ['FNumber', 'ISO', 'GPSLatitude', 0x0004] // 0x0004 is GPSLongitude
}
// is a shortcut for
{exif: ['FNumber', 'ISO'], gps: ['GPSLatitude', 0x0004]}
// which is another shortcut for
{exif: {pick: ['FNumber', 'ISO']}, gps: {pick: ['GPSLatitude', 0x0004]}}

Chunked reader

options.chunked

Type: bool Default: true

Exifr can read only a few chunks instead of the whole file. It's much faster, saves memory and unnecessary disk reads or network fetches. Works great with complicated file structures - .tif files may point to metadata scattered throughout the file.

How it works: A first small chunk (of firstChunkSize) is read to determine if the file contains any metadata at all. If so, reading subsequent chunks (of chunkSize) continues until all requested segments are found or until chunkLimit is reached.

Supported inputs: Chunked is only effective with Blob, <img> element, string url, disk path, or base64. These inputs are not yet processed or read into memory. Each input format is implemented in a separate file reader class. Learn more about file readers and modularity here.

If you use URL as input: Fetching chunks (implemented in UrlFetcher) from web server uses HTTP Range Requests. Range request may fail if your server does not support ranges, if it's not configured properly or if the fetched file is smaller than the first chunk size. Test your web server or disable chunked reader with {chunked: false} when in doubt.

options.firstChunkSize

Type: number Default: 512 Bytes in Node / 65536 (64 KB) in browser

Size (in bytes) of the first chunk that probes the file for traces of exif or metadata.

In browser, it's usually better to read just a larger chunk in hope that it contains the whole EXIF (and not just the beginning) instead of loading multiple subsequent chunks. Whereas in Node.js it's preferable to read as little data as possible and fs.read() does not cause slowdowns.

options.chunkSize

Type: number Default: 65536 Bytes (64 KB)

Size of subsequent chunks that may be read after the first chunk.

options.chunkLimit

Type: number Default: 5

Max amount of subsequent chunks allowed to read in which exifr searches for data segments and blocks. I.e. failsafe that prevents from reading the whole file if it does not contain all of the segments or blocks requested in options.

This limit is bypassed if multi-segment segments ocurs in the file and if options.multiSegment allows reading all of them.

If the exif isn't found within N chunks (64*5 = 320KB) it probably isn't in the file and it's not worth reading anymore.

options.httpHeaders

Type: object Default: {}

Additional HTTP headers to include when fetching chunks from URLs that require Authorization or other custom headers.

Output format

options.mergeOutput

Type: bool Default: true

Merges all parsed segments and blocks into a single object.

Warning: mergeOutput: false should not be used with translateKeys: false or when parsing both ifd0 (image) and ifd1 (thumbnail). Tag keys are numeric, sometimes identical and may collide.

options.sanitize

Type: bool Default: true

Cleans up unnecessary, untransformed or internal tags (IFD pointers) from the output.

options.silentErrors

Type: bool Default: true

Suppresses errors that occur during parsing. Messages are stored at output.errors instead of throwing and causing promise rejection.

NOTE: Some fundamental error's can still be thrown. Such as wrong arguments or Unknown file format.

Failing silently enables reading broken files. But only file-structure related errors are caught.

options.translateKeys

Type: bool Default: true

Translates tag keys from numeric codes to understandable string names. I.e. uses Model instead of 0x0110. Most keys are numeric. To access the Model tag use output.ifd0[0x0110] or output.ifd0[272] Learn more about dictionaries.

Warning: translateKeys: false should not be used with mergeOutput: false. Keys may collide because ICC, IPTC and TIFF segments use numeric keys starting at 0.

options.translateValues

Type: bool Default: true

Translates tag values from raw enums to understandable strings. Learn more about dictionaries.

options.reviveValues

Type: bool Default: true

Converts dates from strings to a Date instances and modifies few other tags to a more readable format. Learn more about dictionaries.

Advanced APIs

Tips for advanced users. You don't need to read further unless you're into customization and bundlers.

This is mostly relevant for Web Browsers, where file size and unused code elimination is important.

The library's functionality is divided into four categories.

• (Chunked) File reader reads different input data structures by chunks. BlobReader (browser), UrlFetcher (browser), FsReader (Node.js), Base64Reader See src/file-readers/. NOTE: Everything can read everything out-of-the-box as a whole file. But file readers are needed to enable chunked mode.
• File parser looks for metadata in different file formats .jpg, .tiff, .heic See src/file-parsers/.
• Segment parser extracts data from various metadata formats (JFIF, TIFF, XMP, IPTC, ICC) TIFF/EXIF (IFD0, EXIF, GPS), XMP, IPTC, ICC, JFIF See src/segment-parsers/.
• Dictionary affects the way the parsed output looks. See src/dicts/.

Each reader, parser and dictionary is a separate file that can be used independently. This way you can configure your own bundle with only what you need, eliminate dead code and save tens of KBs of unused dictionaries.

Check out examples/custom-build.js.

Scenario 1: We'll be handling .jpg files in blob format and we want to extract ICC data in human-readable format. For that we'll need dictionaries for ICC segment.

// Core bundle has nothing in it
import * as exifr from 'exifr/src/core.mjs'
// Now we import what we need
import 'exifr/src/file-readers/BlobReader.mjs'
import 'exifr/src/file-parsers/jpeg.mjs'
import 'exifr/src/segment-parsers/icc.mjs'
import 'exifr/src/dicts/icc-keys.mjs'
import 'exifr/src/dicts/icc-values.mjs'

Scenario 2: We want to parse .heic and .tiff photos, extract EXIF block (of TIFF segment). We only need the values to be translated. Keys will be left untranslated but we don't mind accessing them with raw numeric keys - output[0xa40a] instead of output.Sharpness. Also, we're not importing any (chunked) file reader because we only work with Uint8Array data.

import * as exifr from 'exifr/src/core.mjs'
import 'exifr/src/file-parsers/heic.mjs'
import 'exifr/src/file-parsers/tiff.mjs'
import 'exifr/src/segment-parsers/tiff.mjs'
import 'exifr/src/dicts/tiff-exif-values.mjs'

EXIF Data are mostly numeric enums, stored under numeric code. Dictionaries are needed to translate them into meaningful output. But they take up a lot of space (40 KB out of full build's 60 KB). So it's a good idea to make your own bundle and shave off the dicts you don't need.

• Key dict translates object keys from numeric codes to string names (output.Model instead of output[0x0110])
• Value dict translates vales from enum to string description (Orientation becomes 'Rotate 180' instead of 3)
• Reviver further modifies the value (converts date string to an instance of Date)

Exifr's dictionaries are based on exiftool.org. Specifically these: TIFF (EXIF & GPS), ICC, IPTC, JFIF

// Modify single tag's 0xa409 (Saturation) translation
import exifr from 'exifr'
let exifKeys   = exifr.tagKeys.get('exif')
let exifValues = exifr.tagValues.get('exif')
exifKeys.set(0xa409, 'Saturation')
exifValues.set(0xa409, {
  0: 'Normal',
  1: 'Low',
  2: 'High'
})

// Modify single tag's GPSDateStamp value is processed
import exifr from 'exifr'
let gpsRevivers = exifr.tagRevivers.get('gps')
gpsRevivers.set(0x001D, rawValue => {
  let [year, month, day] = rawValue.split(':').map(str => parseInt(str))
  return new Date(year, month - 1, day)
})

// Create custom dictionary for GPS block
import exifr from 'exifr'
exifr.createDictionary(exifr.tagKeys, 'gps', [
  [0x0001, 'LatitudeRef'],
  [0x0002, 'Latitude'],
  [0x0003, 'LongitudeRef'],
  [0x0004, 'Longitude'],
])

// Extend existing IFD0 dictionary
import exifr from 'exifr'
exifr.createDictionary(exifr.tagKeys, 'ifd0', [
  [0xc7b5, 'DefaultUserCrop'],
  [0xc7d5, 'NikonNEFInfo'],
  ...
])

Parcel works out of the box and Webpack should too because of webpackIgnore magic comment added to the library's source code import(/* webpackIgnore: true */ 'fs').

If this does not work for you, try adding node: {fs: 'empty'} and target: 'web' or target: 'webworker' to your Webpack config. Or similar settings for your bundler of choice.

Alternatively, create your own bundle around core build and do not include FsReader in it.

Exifr is written using modern syntax, mainly async/await. You may need to add regenerator-runtime or reconfigure babel.

Performance

Tips for better performance

Here are a few tips for when you need to squeeze an extra bit of speed out of exifr when processing a large amount of files. Click to expand.

// do this:
let {ISO, FNumber} = await exifr.parse(file, {exif: ['ISO', 'FNumber']})
// not this:
let {ISO, FNumber} = await exifr.parse(file)

// do this:
let options = {ifd0: false, exif: true} 
// not this:
let options = {exif: true} 

// do this:
exifr.gps(file)
// not this:
exifr.parse(file, {gps: true})

// do this:
let options = {exif: true, iptc: true}
for (let file of files) exif.parse(file, options)
// not this:
for (let file of files) exif.parse(file, {exif: true, iptc: true})

Remarks

File reading: You don't need to read the whole file and parse through a MBs of data. Exifr takes an educated guess to only read a small chunk of the file where metadata is usually located. Each platform, file format, and data type is approached differently to ensure the best performance.

Finding metadata: Other libraries use brute force to read through all bytes until 'Exif' string is found. Whereas exifr recognizes the file structure, consisting of segments (JPEG) or nested boxes (HEIC). This allows exifr to read just a few bytes here and there, to get the offset and size of the segment/box and pointers to jump to the next.

HEIC: Simply finding the exif offset takes 0.2-0.3ms with exifr. Compare that to exif-heic-js which takes about 5-10ms on average. Exifr is up to 30x faster.

Benchmarks

pigallery2 did a few benchmarks.

2036 photos (in total 22GB):
lib        | average  | all files
---------------------------------
exifr      | 2.5ms    | 5s         <--- !!!
exifreader | 9.5ms    | 19.5s
exiftool   | 76ms     | 154s

Try the benchmark yourself at benchmark/chunked-vs-whole.js

user reads file            8.4 ms
exifr reads whole file     8.2 ms
exifr reads file by chunks 0.5 ms  <--- !!!
only parsing, not reading  0.2 ms  <--- !!!

Observations from testing with +-4MB pictures (Highest quality Google Pixel photos. Tested on a mid-range dual-core i5 machine with SSD).

• Node: Parsing after fs.readFile = 0.3ms
• Node: Reading & parsing by chunks = 0.5ms
• Browser: Processing ArrayBuffer = 3ms
• Browser: Processing Blob = 7ms
• Browser: <img> with Object URL = 3ms
• Drag-n-dropping gallery of 100 images and extracting GPS data takes about 65ms.
• Phones are about 4x slower. Usually 4-30ms per photo.

Be sure to visit the exifr playground or benchmark/gps-dnd.html, drop in your photos and watch the parsed in timer.

Changelog

For full changelog visit CHANGELOG.md.

Notable changes

• 7.0.0 string URLs as file argument are now accepted in Node.js (UrlFetcher uses polyfill for fetch() in Node.js). But only in full bundle. Breaking change in XMP parsing. Only affects obscure cases with lists and nested rdf:Description, but breaking change nonetheless.
• 6.3.0 AVIF support.
• 6.0.0 & 6.2.0 PNG support.
• 4.3.0 Package.json's "main" now points to UMD bundle for better compatibility.
• 4.1.0 Started bundling shims and polyfills with legacy builds. Suppporting IE10.
• 4.0.0 Added XMP Parser and XMP Extended support.
• 3.0.0 Major rewrite, added ICC parser, HEIC file support, IE11 back compat, reimplemented chunked reader.

F.A.Q.

TL;DR: Because exifr comes in three bundles, each in three format variants (ESM, UMD, legacy), each in two extensions (.js and .mjs or .mjs) due to tooling. Plus source codes are included.

npm (~1MB, ~65 files): The module includes both src/ and dist/. That's source codes of all the readers, parsers and dictionaries. Multiplied by 3 bundles (full, lite, mini). Then multiplied by 3 bundle formats (ESM, UMD, legacy for IE10) and multiplied by 2 extensions (.mjs+.js or .cjs+.js). But you won't use all of the files. They're there so you can choose what's best for your project, tooling and environment.

bundlephobia (~63/22 kB): Full build is the "main" entry point (in package.json) picked up by Node and bundlephobia. But it's meant for use in Node where size doesn't matter.

badge in readme (~9 kB): The badge points to mini bundle which contains the bare minimum needed to cover the most use-cases (get orientation, coords, exif info, etc...). This is meant for browsers where file size matters.

Contributing

Contributions are welcome in any form. Suggestions, bug reports, docs improvements, new tests or even feature PRs. Don't be shy, I don't bite.

If you're filing an issue, please include:

• The photo that's missing metadata or causing the bug
• Repo or a sandbox (like this one) with minimal code where the bug is reproducible.

There are so many environments, tools and frameworks and I can't know, nor try them all out. Letting me peek into your setup makes tracking down the problem so much easier.

PRs are gladly accepted. Please run tests before you create one:

• in browser by visiting /test/index.html (uses import maps, you may need to enable experimental flags in your browser)
• in Node.js by running npm run test

License

MIT, Mike Kovařík, Mutiny.cz