Is the Fervor for Generative AI Cooling?

Or is it all just a hallucination?

Hello and thank you for subscribing to the Photo AI newsletter! I’ve been mighty busy lately, hence the span since the last piece was published. Want this to happen more frequently? Consider becoming a paid subscriber to support my work for as little as $4.17 per month. I could make a joke about how that’s less than buying me a cup of coffee, but I take coffee seriously. For the folks who already support this newsletter financially, thank you! And if you aren’t contributing monetarily, I appreciate your time and attention.

First Thing

I’ve just released a big new third edition of my book Take Control of Your Digital Photos! It was last updated in 2019, and as you can imagine, the photography field has changed quite a bit in the last four years. (Have you heard of this new thing called AI??) Heck, photography has changed a lot in the last ten years, which is when (almost to the month) Take Control Books published the first edition.

And yet, we’re still drowning in photos, even more than ever. Fortunately, many new and different methods of dealing with the flood can help you keep your head above water. Don’t like assigning keywords to photos? Machine learning technologies can do it for you, as with the utilities Excire Foto and ON1 Photo Keyword AI, or you can rely on what some apps, like Apple Photos and Lightroom desktop, come up with on their own when you perform a search. Do you feel like you’ll never get ahead of reviewing your images to find the great photos nestled in the pile? This book will help.

Most of all, Take Control of Your Digital Photos gives you a plan and clear workflows for organizing your photo library in a sensible way that respects your time and effort. The book makes it easier to find your best images for editing or sharing with friends and family—or just for you to enjoy the photos you worked so hard to capture! I’ve added entirely new sections on coming up with a workflow, working with multiple photo libraries, scanning for metadata, and more.

Take Control of Your Digital Photos is available now as a 184 page ebook—that’s 20 new pages of content since the last edition, even after removing a few outdated sections—for just $14.99. If you own an earlier version, log in to your Take Control account for special upgrade pricing of just $5!

Is the Fervor for Generative AI Cooling?

This newsletter is all about how AI and ML technologies are fundamentally changing photography, and in recent months it’s felt like Generative AI in particular has dominated not just the conversation but pop culture. At the beginning of the year I wondered Can We Escape GenAI?, and mused about AI-Generated Imagery and Creative Ownership when I made my own entirely GenAI creations. It’s super cool to type the words of an idea and have that rendered in photorealistic imagery.

But I also wonder if the excitement over Generative AI is starting to wane. Or at the very least, people are seeing the results and realizing that, although the technology is impressive, the results are still largely falling short of what we expect them to be.

A big reason for this is the way generative systems “hallucinate.” That’s the wonderful term that’s stuck to describe when an AI system makes up wrong answers or comes to nonsensical conclusions. (Another word for this is “failure.”) This is happening more in LLMs (large language models) like ChatGPT, which can turn outright deadly, such as AI-generated books about plant and mushroom foraging that incorrectly identify species that are deadly (see “‘Life or Death:’ AI-Generated Mushroom Foraging Books Are All Over Amazon” at 404 Media).

But it seems to also be happening in imagery, too. I’m seeing fewer instances of people creating fully-AI-generated images. Instead, I’m running across more examples of the technology being employed as tools. Using Generative Fill in the Photoshop beta to remove items and clean up areas of a photo is truly groundbreaking, especially if you’ve had to spend time erasing areas larger than dust spots in images. Apps like Photomator are adding all sorts of helpful ML-based enhancements that leverage the underlying technology in narrow, practical ways; even Capture One Pro, which seemed like it would be skipping the ML revolution, now includes Auto Dust Removal, Smart Adjustments across images, and Face Focus to check sharpness of faces while culling images.

At the very least, people are realizing that getting high-quality results takes a lot of time, iteration, and patience. Sure, you can ask it to create a mermaid in an area of your photo, but on the first try you probably won’t like what you see. You need to experiment with other generative AI systems, more descriptive prompts, and so on (I did a quick attempt on a photo of a waterfall using Photoshop here).

The original photo (left); with the prompt “mermaid with red hair” (middle); and with the prompt “cairn of rocks and reflection” (right).

To be fair, my impression could be due to self-selected algorithmic effects. I’m not searching them out, so the various services that feed me new images just aren’t showing a lot of Generative AI versions to me. Or perhaps this is me projecting my own wish fulfillment onto the field. For photography, even though amazing results are possible, I see specific tools that harness AI capabilities as the more likely way forward.

Podcasts

New episodes of PhotoActive are out!

• Episode 148: Creativity with S J Watson: “With writing I’m trying to use a story to create images, and with photography I’m trying to use images to create a story.” That’s S J Watson, best-selling novelist of Before I Go to Sleep. In this episode we talk to him about how street photography feeds his writing and vice-versa, and the way different endeavors contribute to creative mindsets.

• Episode 149: Daniel Agee on Glass Year Two: In episode 111, we talked to the founders of the photo app and community Glass about their approach to making a social photo service that isn’t exhausting or frustrating like Instagram or many others. For this episode, we catch up with Daniel Agee, the head of Marketing and Community at Glass, to talk about Glass’s two year anniversary, eschewing VC funding, and how community is making such a difference.

Further

• Why isn’t camera image quality improving as fast as smartphone IQ?, Richard Butler at DPReview

• Federal Judge Rules AI Images Cannot be Copyrighted, Contrasts it to Photos

Let’s Connect

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.