exiftool-vendored

Fast, cross-platform Node.js access to ExifTool.

Features

1. Best-of-class cross-platform performance and reliability.

   Expect an order of magnitude faster performance than other packages.

2. Proper extraction of

   • dates with correct timezone offset encoding, when available)
   • latitudes & longitudes as floats (where negative values indicate west or south of the meridian)

3. Support for

   • reading tags
   • extracting embedded binaries, like thumbnail and preview images
   • writing tags
   • rescuing metadata

4. Robust type definitions of the top 99.5% tags used by over 6,000 different camera makes and models (see an example)

5. Automated updates to ExifTool (as new versions come out monthly)

6. Robust test coverage, performed with on macOS, Linux, and Windows

Installation

 yarn add exiftool-vendored

or

 npm install --save exiftool-vendored

Note that exiftool-vendored provides an installation of ExifTool relevant for your local platform through optionalDependencies.

You shouldn't include either the exiftool-vendored.exe or exiftool-vendored.pl as direct dependencies to your project.

If you're installing on a minimal Linux distribution, like Alpine, you may need to install perl (à la "apk add perl").

Upgrading

See the CHANGELOG for breaking changes since you last updated.

Major version bumps

I bump the major version if there's a chance existing code might be affected.

I've been bit too many times by my code breaking when I pull in minor or patch upgrades with other libraries. I think it's better to be pessimistic in code change impact analysis: "over-promise and under-deliver" your breaking-code changes.

When you upgrade to a new major version, please take a bit more care in validating your own systems, but don't be surprised when everything still works.

Usage

There are many configuration options to ExifTool, but all values have (more or less sensible) defaults.

Those defaults have been used to create the exiftool singleton. Note that if you don't use the default singleton, you don't need to .end() it.

// We're using the singleton here for convenience:
const exiftool = require("exiftool-vendored").exiftool

// And to verify everything is working:
exiftool
  .version()
  .then((version) => console.log(`We're running ExifTool v${version}`))

If the default ExifTool constructor parameters wont' work for you, it's just a class that takes an options hash:

const ExifTool = require("exiftool-vendored").ExifTool
const exiftool = new ExifTool({ taskTimeoutMillis: 5000 })

You should only use the exported default exiftool singleton, or only create one instance of ExifTool as a singleton.

Remember to .end() whichever singleton you use.

General API

ExifTool.read() returns a Promise to a Tags instance. Note that errors may be returned either by rejecting the promise, or for less severe problems, via the errors field.

All other public ExifTool methods return Promise<void>, and will reject the promise if the operation is not successful.

Tags types

ExifTool knows how to extract several thousand different tag fields.

Unfortunately, TypeScript crashes with error TS2590: Expression produces a union type that is too complex to represent if the Tags interface was comprehensive.

Instead, we build a corpus of "commonly seen" tags from over 10,000 different digital camera makes and models, many from the ExifTool metadata repository and <raw.pixls.us>.

Here are some example fields:

  /** ★☆☆☆ ✔ Example: 200 */
  ISO?: number

  /** ★★★★ ✔ Example: 1920 */
  ImageHeight?: number

  /** ★★★★ ✔ Example: 1080 */
  ImageWidth?: number

  /** ★★★★ ✔ Example: "image/jpeg" */
  MIMEType?: string

The stars represent how common that field has a value in the example corpus. ★★★★ fields are found in > 50% of the examples. ☆☆☆☆ fields are found in < 1% of examples.

The checkmark denotes if the field is found in "popular" cameras (like recent Nikon, Canon, Sony, and Apple devices).

Caveats with Tags

The fields in Tags are not comprehensive.

Just because a field is missing from the Tags interface does not mean the field doesn't exist in the returned object. This library doesn't exclude unknown fields, in other words. It's up to you and your code to look for other fields you expect and cast to a more relevant interface.

Logging and events

To enable trace, debug, info, warning, or error logging from this library and the underlying batch-cluster library, provide a Logger instance to the ExifTool constructor options.

ExifTool instances emits many lifecycle and error events via batch-cluster.

Reading tags

exiftool
  .read("path/to/image.jpg")
  .then((tags /*: Tags */) =>
    console.log(
      `Make: ${tags.Make}, Model: ${tags.Model}, Errors: ${tags.errors}`
    )
  )
  .catch((err) => console.error("Something terrible happened: ", err))

Extracting embedded images

Extract the low-resolution thumbnail in path/to/image.jpg, write it to path/to/thumbnail.jpg, and return a Promise<void> that is fulfilled when the image is extracted:

exiftool.extractThumbnail("path/to/image.jpg", "path/to/thumbnail.jpg")

Extract the Preview image (only found in some images):

exiftool.extractPreview("path/to/image.jpg", "path/to/preview.jpg")

Extract the JpgFromRaw image (found in some RAW images):

exiftool.extractJpgFromRaw("path/to/image.cr2", "path/to/fromRaw.jpg")

Extract the binary value from "tagname" tag in path/to/image.jpg and write it to dest.bin (which cannot exist already and whose parent directory must already exist):

exiftool.extractBinaryTag("tagname", "path/to/file.exf", "path/to/dest.bin")

Writing tags

Note that only a portion of tags is writable. Refer to the documentation and look under the "Writable" column.

If you apply malformed values or ask to write to tags that aren't supported, the returned Promise will be rejected.

Only string and numeric primitives are supported as values to the object.

To write a comment to the given file so it shows up in the Windows Explorer Properties panel:

exiftool.write("path/to/file.jpg", { XPComment: "this is a test comment" })

To change the DateTimeOriginal, CreateDate and ModifyDate tags (using the AllDates shortcut) to 4:56pm UTC on February 6, 2016:

exiftool.write("path/to/file.jpg", { AllDates: "2016-02-06T16:56:00" })

To write to a specific metadata group's tag, just prefix the tag name with the group. (TypeScript users: you'll need to cast to make this compile).

exiftool.write("path/to/file.jpg", {
  "IPTC:CopyrightNotice": "© 2021 PhotoStructure, Inc.",
})

To delete a tag, use null as the value.

exiftool.write("path/to/file.jpg", { UserComment: null })

The above example removes any value associated with the UserComment tag.

Always Beware: Timezones

If you edit a timestamp tag, realize that the difference between the changed timestamp tag and the GPS value is used by exiftool-vendored to infer the timezone.

In other words, if you only edit the CreateDate and don't edit the GPS timestamps, your timezone will either be incorrect or missing. See the section about Dates below for more information.

Rewriting tags

You may find that some of your images have corrupt metadata and that writing new dates, or editing the rotation information, for example, fails. ExifTool can try to repair these images by rewriting all the metadata into a new file, along with the original image content. See the documentation for more details about this functionality.

rewriteAllTags returns a void Promise that will be rejected if there are any errors.

exiftool.rewriteAllTags("problematic.jpg", "rewritten.jpg")

ExifTool configuration support (.ExifTool_config)

ExifTool has an extensive user configuration system. There are several ways to use one:

1. Place your user configuration file in your HOME directory
2. Set the EXIFTOOL_HOME environment variable to the fully-qualified path that contains your user configuration.
3. Specify the in the ExifTool constructor options:

new ExifTool({ exiftoolEnv: { EXIFTOOL_HOME: resolve("path", "to", "config", "dir") }

This package requires procps.

The default BatchClusterOptions.cleanupChildProcs value of true means that BatchCluster, which is used to manage child exiftool processes, will try to use ps to ensure Node's view of process state are correct, and that errant processes are cleaned up.

If you run this in a docker image based on Alpine or Debian Slim, this won't work properly unless you install the procps package.

See batch-cluster for details.

Resource hygiene

Call ExifTool.end() when you're done

You must explicitly call .end() on any used instance of ExifTool to allow node to exit gracefully.

ExifTool child processes consume system resources, and prevents node from exiting due to the way Node.js streams work.

Note that you can't call cannot be in a process.on("exit") hook, as the stdio streams attached to the child process cannot be unref'ed. (If there's a solution to this, please post to the above issue!)

Mocha v4.0.0

If you use mocha v4 or later, and you don't call exiftool.end(), you will find that your test suite hangs. The relevant change is described here, and can be solved by adding an after block that shuts down the instance of ExifTool that your tests are using:

after(() => exiftool.end()) // assuming your singleton is called `exiftool`

Dates

The date metadata in all your images and videos are, most likely, underspecified.

Images and videos rarely specify a time zone in their dates. If all your files were captured in your current time zone, defaulting to the local time zone is a safe assumption, but if you have files that were captured in different parts of the world, this assumption will not be correct. Parsing the same file in different parts of the world result in different times for the same file.

Prior to version 7, heuristic 1 and 3 were applied.

As of version 7.0.0, exiftool-vendored uses the following heuristics. The highest-priority heuristic to return a value will be used as the timezone offset for all datetime tags that don't already have a specified timezone.

Heuristic 1: explicit metadata

If the EXIF TimeZoneOffset tag is present it will be applied as per the spec to DateTimeOriginal, and if there are two values, the ModifyDate tag as well. OffsetTime, OffsetTimeOriginal, and OffsetTimeDigitized are also respected, if present (but are very rarely set).

Heuristic 2: GPS location

If GPS latitude and longitude is present and valid (the value of 0, 0 is considered invalid), the tz-lookup library will be used to determine the time zone name for that location.

Heuristic 3: UTC timestamps

If GPSDateTime or DateTimeUTC is present, the delta with the dates found within the file, as long as the delta is valid, is used as the timezone offset. Deltas of > 14 hours are considered invalid.

ExifDate and ExifDateTime

Because date-times have this optionally-set timezone, and some tags only specify the date, this library returns classes that encode the date, the time of day, or both, with an optional timezone and an optional tzoffset: ExifDateTime and ExifTime. It's up to you, then, to determine what's correct for your situation.

Note also that some smartphones record timestamps with microsecond precision (not just milliseconds!), and both ExifDateTime and ExifTime have floating point milliseconds.

Tags

Official EXIF tag names are PascalCased, like AFPointSelected and ISO. ("Fixing" the field names to be camelCase, would result in ungainly aFPointSelected and iSO atrocities).

The Tags interface is auto-generated by the mktags script, which parses through over 6,000 unique camera make and model images, in large part sourced from the ExifTool site. mktags groups tags, extracts their type, popularity, and example values such that your IDE can autocomplete.

Tags marked with "★★★★", like MIMEType, should be found in most files. Of the several thousand metadata tags, realize less than 50 are found generally. You'll need to do your research to determine which tags are valid for your uses.

Note that if parsing fails (for, example, a date-time string), the raw string will be returned. Consuming code should verify both existence and type as reasonable for safety.

Serialization

The Tags object returned by .readTags() can be serialized to JSON with JSON.stringify.

To reconstitute, use the parseJSON() method.

import { exiftool, parseJSON } from "exiftool-vendored"

const tags: Tags = await exiftool.readTags("/path/to/file.jpg")
const str: string = JSON.stringify(tags)

// parseJSON doesn't validate the input, so we don't assert that it's a Tags
// instance, but you can cast it (unsafely...)

const tags2: Tags = parseJSON(str) as Tags

Performance

The default exiftool singleton is intentionally throttled. If full system utilization is acceptable:

1. set maxProcs higher

2. consider setting minDelayBetweenSpawnMillis to 0

3. On a performant linux box, a smaller value of streamFlushMillis may work as well: if you see noTaskData events, you need to bump the value up.

Benchmarking

The yarn mktags ../path/to/examples target reads all tags found in a directory hierarchy of sample images and videos, and parses the results.

exiftool-vendored v16.0.0 on a 2019 AMD Ryzen 3900X running Ubuntu 20.04 on an SSD can process 20+ files per second per thread, or 500+ files per second when utilizing all CPU threads.

Batch mode

Using ExifTool's -stay_open batch mode means we can reuse a single instance of ExifTool across many requests, dropping response latency dramatically as well as reducing system load.

Parallelism

To avoid overwhelming your system, the exiftool singleton is configured with a maxProcs set to a quarter the number of CPUs on the current system (minimally 1); no more than maxProcs instances of exiftool will be spawned. If the system is CPU constrained, however, you may want a smaller value. If you have very fast disk IO, you may see a speed increase with larger values of maxProcs, but note that each child process can consume 100 MB of RAM.

Author

• Matthew McEachen

Contributors 🎉

• Joshua Harris
• Anton Mokrushin
• Luca Ban
• Demiurga
• David Randler

CHANGELOG

See the CHANGELOG on github.