Here’s the problem with keywords: although they make finding photos easier, the act of tagging images is time-consuming busywork we don’t want to do. So we plod forward, skipping the keywording step and relying on our fuzzy memories to scan through libraries looking for the images we want. But what if we could get descriptive keywords without entering them ourselves? That’s the promise of ON1 Photo Keyword AI, a new utility that uses AI technologies to identify scenes and objects in images and create relevant keywords.
The big players have turned to AI to bypass the keywording stage, but in an often frustrating, opaque way. Apple Photos, Google Photos, and Adobe Lightroom (the desktop and mobile versions, not Lightroom Classic), for example, all scan your images for things the machine learning models recognize. The upside is that you can search for things like “leaves” or “cloudy skies” or “cars” and usually get results that contain the items, even if the images were never specifically tagged with those terms.
The downside is that you don’t know which objects or characteristics are associated with any given photo. You’re going on faith that the app or service is doing a good job. Did the software grab every image in your library that contains a car? What about less tangible terms? A recent search for “snow” in Apple Photos brought up snowy scenes, but also a black and white photo and an image of pale rock formations in the summer.
|The Apple Photos app doesn’t know that the black and white photo or the rocks at right do not actually contain snow.|
ON1 Photo Keyword AI does the same type of scanning, but presents the actual keywords it generated and adds them to photos in a transparent way.
How ON1 Photo Keyword AI works
ON1 Photo Keyword AI is a standalone app built on ON1’s browsing and catalog technology that undergirds its flagship ON1 Photo RAW editor. (In fact, the latest release, ON1 Photo RAW 2023.5, incorporates the Photo Keyword AI features.) Unlike the company’s other individual tools, this one is not available as a plug-in for other apps such as Lightroom Classic, though there is a way to make it work with other apps (which we’ll get to later).
ON1 Photo Keyword AI is a standalone purchase retailing for $69.99, which includes activation on up to two computers, and is available for macOS and Windows systems.
To get started in ON1 Photo Keyword AI, you browse to a folder of images on disk, and then select one or more photos to scan.
|Browse a folder of images in ON1 Photo Keyword AI.|
In the Metadata panel at right, expand the AI Keywords section (if it’s not visible) and click the Scan button. The app reviews the image(s) and builds a set of keywords that appear in the field below.
|After scanning, ON1 Photo Keyword AI has come up with a set of keywords based on what it identified in the selected image. (“Cascades2” is the name of the folder in which the photo appears.)|
It’s important to point out that the scanning happens locally; images are not uploaded to a cloud processor or used for further machine learning training. The scanning time depends on the capabilities of your computer and the number of images selected for scanning. On a 2021 MacBook Pro with an M1 Max processor and 32 GB of memory, scanning a single image took a few seconds. Processing a folder of 74 raw images took just under 6 minutes. A batch of 500 photos from an event containing numerous people took about 38 minutes.
By default, the generated keywords are automatically applied to the photos, either embedded directly into the image files (for formats such as JPEG) or written to .XMP sidecar files (for raw images). You can see the terms added in the Keywords field.
It’s important to point out that the scanning happens locally; images are not uploaded to a cloud processor or used for further machine learning training.
Alternatively, you can choose to add the keywords manually by turning off the Automatically Embed Metadata setting. In that case, clicking the arrow icon on a suggested keyword adds it to the selected image(s). (Clicking the X that appears when you mouse over a suggested keyword removes it from the AI Keywords field.)
Or, if you’re satisfied with the results of a scan, you can click Add All to tag the photo(s) with every found keyword. Be aware, though, that if multiple images are selected, they each get all the terms, even ones that were not suggested for specific photos. For example, if a person appears in one image but not another, and they’re both selected, clicking Add All tags both photos with the “Person” keyword.
In addition to working with photos on your drives, ON1 Photo Keyword AI can scan images as you import them from a camera or memory card, front-loading the keyword process in your workflow. They get saved to a folder of your choice and tagged with the discovered keywords; if you use an app for organizing your library that recognizes XMP files, the keywords should appear when you add the photos.
|ON1 Photo Keyword AI can import photos directly from a card or camera and scan for keywords during ingest.|
One of the values of an app like this is the ability to make suggestions you may not consider. For instance, a photo of a person taking a photo of trees brings up descriptors such as “Outerwear,” “Backpack,” and “Luggage and bags” in addition to expected terms like “Person” and “Tree.” It also pulls words from metadata such as the location-based “Stehekin” (a town name) and “Chelan County” where the image was captured, if the GPS data is already embedded in the image file.
|The app may generate more keywords than needed, but it also incorporates information like location (“Chelan County” here) if present in the file.|
That said, the app does tend to throw everything it finds at that Keyword field. Is it important that a series of photos contain the term “Biome” or “Habitat”? Probably not. In our testing it also tended to mark many animals as “Bear” and added “Carnivore” to the mix. Your pet may indeed be a carnivore, but that doesn’t seem like a keyword you’d often search for when trying to locate images of your cat FluffySnuggles. (Counterpoint: cats.) As another example, we also ran instances where scanning portraits and photos specifically including people served up keywords such as “Human arm,” “Human hair,” “Finger,” “Organ (Biology),” and “Human action.”
It’s easy to make fun of such results because our brains automatically sort out unnecessary terms when we’re evaluating photos. The machine learning models are designed to describe the contents of scenes based on what the models have been exposed to previously. Honestly, it’s probably better to include too many keywords than too few.
And some of those terms are helpful, such as marking images that have “1 Face” or “2 Faces.” The app does a good job of estimating general ages, making it possible to identify photos that contain “Child,” “Teenager/Young Adult,” or “Elderly” subjects. However, although it can discern when people are in the frame, the app isn’t doing any person recognition. It would be helpful to tag all selected images that contain your friends John or Clara, for instance.
If that’s too much, the ON1 Photo Keyword AI settings include general categories that can be turned on or off, such as Photographic Properties and Histogram Properties.
|Choose which broad categories of information to include during a scan (default selections shown).|
For terms the software didn’t generate, you can type new keywords and append them to the selected images. You can also remove terms you don’t want, although with the default settings it’s a two-step affair: you must first remove the term from the AI Keywords field (by clicking the X that appears on its icon) and then deleting it from the Keywords field. You can scroll down through the Keyword List, which is the database of all keywords collected or generated, and deselect or delete terms, but that’s a lot of steps.
Interacting with other apps
If you use another app to manage your photo library, you have a few options for integrating ON1 Photo Keyword AI into your workflow. One approach would be to import photos from the camera or memory card directly into ON1 Photo Keyword AI (or copy them to your drive and open them in the app), scan for keywords, and then import the processed files in your library manager.
Or, if the photos are already in your library, process the files using ON1 Photo Keyword AI and synchronize the metadata in the library app. For example, you can navigate to a folder that Lightroom Classic already tracks, and then scan the images, which by default embeds the keywords. Lightroom tends to be cautious about blindly updating files it already watches, so you need to manually synchronize the images’ metadata with what’s on disk.
When you see the icon on photos indicating that the metadata on disk is different from what Lightroom expects, click it and choose to overwrite the metadata from disk.
|After reading the updated metadata from disk, Lightroom Classic includes the keywords in its Keywording panel. (Terms with asterisks appear in some, not all, of the selected images.)|
In Capture One, as another example, you’d select the images that were processed in ON1 Photo Keyword AI, right-click, and choose Sync Metadata from the contextual menu. Apple Photos, on the other hand, ignores XMP files, so keywords will only appear in photos where the terms have been written to the image file, such as JPEG or DNG images.
Although the AI scanning is the app’s headlining feature, ON1 Photo Keyword AI is also a full metadata manager, including star ratings, flags, color labels, and IPTC data. There’s also a Map view and the ability to compare photos when evaluating how to rate them. Basically, you could do all of your sorting and culling in ON1 Photo Keyword AI before passing the photos along to the image editor of your choice.
The app builds a database of keywords you’ve used and that have been generated, which are editable in the Keyword List. Misspell a keyword you entered? It can be corrected here, which applies the change to any photo the app tracks. (If the photos were previously added to your photo editor, such as Lightroom Classic, you would need to synchronize the metadata again.)
You can also create Cataloged Folders for frequently-used folders (like a portfolio you add images to regularly), which caches the metadata for faster performance within ON1 Photo Keyword AI, reducing the reliance on reading data from every file on disk.
Let the machine do it
Generating keywords seems like a perfect task for AI. It’s coming up with terms, many that you may not have thought to add, and applying them to many images in a short amount of time. Granted, as with any current AI technology, the results that appear after a scan get you most of the way there–you may end up adding some specific terms and deleting others to match what the software may have missed or which are specific to your needs (like moods or event names). The advantage is having many valid terms to work with instead of nothing at all.
In our testing, we found that adding a separate utility dedicated to keywording required us to reevaluate our workflows. Running ON1 Photo Keyword AI and synchronizing the metadata in Lightroom Classic is an extra step beyond our usual approach of adding keywords during the import stage. (If you already use ON1 Photo RAW as your library manager and photo editor, the feature is already built in.) However, having more metadata–particularly terms we don’t have to come up with ourselves–makes the addition worthwhile. And especially if you don’t currently apply any keywords to your photos (but know that you probably should), throwing the app into your workflow ends up giving you plenty of searchable metadata with very little work.
|Even their distinctive pink-hued appearance doesn’t register to the app that these are flamingos.|
If, on the other hand, you’re fastidious about the keywords you apply, ON1 Photo Keyword AI might be overkill, or at least overwhelming. You’ll likely need to add context-specific keywords to the mostly generic terms that the software adds. For instance, in a recent scan ON1 Photo Keyword AI correctly determined that a photo included birds, but wasn’t savvy enough to identify that they were flamingos. But that’s something you would have added anyway without using a separate utility for generating the keywords.