The Paralysis of “Best”
Can AI/ML help create or determine which photo is "best?"
Hello, and thanks for subscribing to Photo AI! If you’re reading this on the web, click here to subscribe for a free or paid subscription, either of which help me as an independent journalist and photographer. As a quick reminder, I’m on Mastodon at @jeffcarlson@twit.social, co-host the podcasts PhotoActive and Photocombobulate, and have written more than 80 books about technology and photography.
First Thing
Aaron Hockley returned from the Imaging USA conference as an in-person attendee and writes about how two aspects of AI-assisted photo processing are no longer on the fringes: I’m Calling It: AI Photo Culling and Editing Are Mainstream.
The Paralysis of “Best”
Last week the good folks over at Glass posted on Mastodon:
Hi. Hit us with your best photo on Glass. Just because.
🗣️ Show 🗣️ us 🗣️ your 🗣️ photos"
I’ve been a Glass user since its beginning and love how it lets photographers focus on their work, not chase an algorithm the way Instagram does. Glass is a paid subscription service, just $4.99 per month or $29.99 per year, which I’ve found to be a solid deal. Personally, I like browsing photos and interacting with the community, because it’s not a stressful activity. I don’t think about whether I’ve included the correct hashtags, or if people will see my photos at all, or be on alert for auto-playing videos and sponsored posts. I can relax and enjoy photos. (Founders Tom Watson and Stefan Borsje joined Kirk McElhearn and I on our PhotoActive podcast to talk about the service last year.)
But back to that post. My first inclination was to start digging through my photo library for top-rated images. Last year was a good photography year for me thanks to several trips, so should I pick something from those collections? Or should I go back further? And which would I say is my best photo? And is that separate from my favorite photo?
And in that flood of questions, I locked up.
The task was too big. It was a deep rabbit hole that I knew I didn’t have the time to fully explore.
But it did make me think of a few warrens worth nosing into in regards to computational photography. (Oh, you knew I’d get around to AI soon enough.) Let’s examine a few.
Finders, Keepers
The first is the mechanics of finding one’s best shots. I’ve long used a star-rating system to rank my photos, both in preparing them for edit and to mark the ones that rise above the others after editing. Some people prefer to simply flag good images as favorites. When I’m looking for photos in my library to share or use as an example, I can quickly filter for 3-star-and-higher images in Lightroom. The Lightroom mobile apps include a Best Photos feature that will scan through any album containing 2000 images or less and algorithmically narrow down what it thinks are the best shots, taking into account your own ratings plus AI assistance to spot people, in-focus shots, and other criteria. (To do this: Tap a collection to view its contents, then tap the three-dots icon (…) and tap Choose Best Photos.) Photoshop Elements even includes something like this: in the Organizer, click the Auto Curate button.
I’m also a big believer in using keywords and people-recognition, which front-loads the search process so when you need to find something later, you’re not starting from scratch. But I acknowledge that most people don’t assign keywords, which is why one of the earlier consumer-facing AI technologies for photographers was object and scene recognition. Apple Photos and Lightroom desktop (the cloud-focused version) bring up all sorts of relevant tasks when you type something into their search fields, even if you didn’t specifically assign tags to them.
The utility Excire Foto (stand-alone) and Excire Search (Lightroom plug-in) uses that same idea and creates textual keywords that you can apply to your images. I wrote about Excire Foto in Popular Photography: Excire Foto 2022 can analyze and keyword your entire photo library using AI.
And lest I forget to mention it, I’ve written an entire book called Take Control of Your Digital Photos that includes strategies for wrangling photo libraries.
Best by What Criteria?
Now we get to the editing side of things, and the more nebulous areas of determining quality. I have decent photographic gear and have been fortunate to put myself into places where I can capture good photos. I also know how to shoot with an edit in mind, because I know my editing tools well.
Those things will give me advantages over photographers who don’t edit their images and who don’t leave their homes or neighborhoods, but…aha! You can probably think of half a dozen examples of photographers who make better photos than I do under those circumstances.
But stick with me here, and let’s just drill down to the editing side of things. Can I make better photos with the help of machine-learning features? I can confidently say I know enough about photo editing that if those AI tools didn’t exist, I could manually edit a photo to improve it. But those tools can do the same work in much less time. For someone who’s just starting out, clicking the Auto button in editing software will yield results that many people had to learn through years of trial and error.
That doesn’t guarantee that AI made the photo a “best” image, though.
One more example: I can use the face masking features of Lightroom, or the Portrait controls in Luminar Neo or ON1 Photo RAW, and do a perfectly good job retouching a portrait. Will it be as good as the same portrait edited by a professional retoucher? Likely no. Will it be good enough for the client? Almost certainly.
In this case, AI editing has allowed me to do more than I could previously, and do it faster, and even do it at a higher quality level than before. But I’m still making the choices of where to adjust light, how much skin smoothing to apply, which colors to enhance or recede, and so forth.
Does AI automatically push my good photos into the “best” category? No.
Best in Show (Don’t Tell)
Part of what brought up the question of best this week was news that an AI-generated image took top prize in a photo competition. In “This AI Image Fooled Judges and Won a Photography Contest”, PetaPixel writes that the company Absolutely AI entered a drone photo depicting sunset light hitting the crest of a wave to be able to say they created “the world’s first AI-generated award-winning photograph.”
Under closer scrutiny, could the judges have determined it was AI-generated? Perhaps. Is it a sneaky way to get publicity? Certainly, which is why we’ll have to be more diligent about enforcing rules about whether GenAI imagery is allowed in circumstances like these.
But is it a good image? Yeah, it is.
More of “Guess the AI Photo”
Speaking of trying to discern photo reality from fiction, a couple people pointed me to this post at PetaPixel about a magazine that asked its readers to guess which cover was a real photo and which was GenAI. I won’t give it away, but I had to look long and hard before making a decision. (They were smart not to include the people’s fingers, which AI has lots of trouble with.)
Imposter Syndrome on Photocombobulate
In the latest episode of the Photocombobulate podcast, Mason Marsh and I face our fears about imposter syndrome. In our previous episode, we talked to the wonderful and talented Julieanne Kost, where she admitted to also feeling like an imposter sometimes. How do we and other photographers overcome feelings of inadequacy, particularly amid the torrent of great photos on social media? Listen in your favorite podcast player or here: Episode 28: Imposter Syndrome. We also offer a video version:
Let’s Talk
Thanks again for reading and recommending Photo AI to others who would be interested. Send any questions, tips, or suggestions for what you’d like to see covered at jeff@jeffcarlson.com. Are these emails too long? Too short? Let me know.