Firefly Is Adobe’s Shiny New GenAI Tool
Now in beta, coming soon to an Adobe app near you no doubt
Hi again, 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, and co-host the podcasts PhotoActive and Photocombobulate.
First Thing
I was shocked to learn that venerable photography site DPReview is being shut down in April by its corporate owner Amazon.com. I’ve written for DPReview since 2017, and all the people I’ve worked with have been top-notch humans with an amazing array of photographic information. I wrote more about the news at TidBITS: Amazon Shutters Photo Resource Website DPReview.
I’m also sad about this because a few assigned articles of mine in the pipeline are now cancelled. So this is also a timely reminder that your support of this newsletter helps me pay the bills and continue writing about the field of photography and computational photography. If you’re not yet a paid subscriber, click below to chip in as little as $5 per month.
Adobe Firefly Beta
Another development last week was the unveiling of Adobe Firefly, the company’s foray into Generative AI (GenAI).
For now the service is very much in beta: you must apply to be on the waitlist, the images exhibit a watermark that specifies they’re for non-commercial use only, and you cannot use your own images as source material. I’m sure the latter feature will arrive sooner rather than later, since there’s a lot of power in being able to generate imagery around existing content. But what I’ve seen so far has been occasionally jaw-dropping (and sometimes horrifying).
I’ll admit that so far I haven’t gotten too deep into Generative AI because it’s involved several steps and complications: get access, join a Discord server, enter text prompts in the right area using the correct syntax, and keep track of how many credits you’ve used to generate images.
Adobe Firefly abstracts that all away with a friendly, simple interface that (at least so far) doesn’t limit how many images you can create. Options such as the type of artwork (photo, art) to generate and the camera view (closeup, wide angle) are set using menus in a sidebar, even though you can also just type those terms into the text prompt at the bottom of the screen.
It’s also more friendly in terms of the content. When you enter a term that isn’t allowed, Firefly pops up a polite notice that it won’t generate an image. However, it doesn’t specify which term was problematic, so you often need to guess. Some are obvious, such as the names of notable people (you can’t use Firefly to make propaganda about the former president, for example, as many people have done lately with other GenAI systems). But I found that the word “battle” was also excluded when I tried to make an image of a fantasy warrior holding a battle axe.
(I also asked it to generate a version with the warrior wearing a helmet and mask and it came up with a result showing proper Covid protocols in place in the fantasy realm. Ha!)
However, sometimes the results are eerily realistic. I made a couple of portraits that look nearly flawless like real photos.
And for landscape photography, unless you’re trying to depict a specific place, the images can be gorgeous.
Adobe will eventually work this technology into its products. The Firefly site includes features that are coming but not yet available, such as recoloring vector lines and shapes. There’s also an option to apply generative effects to text, which is fun.
I’m resisting getting into What It All Means, because we don’t know yet. Did Content-Aware Fill in Photoshop make things worse for photo retouchers? Perhaps for some, but I think it also allowed them to work more quickly on trivial corrections and focus on areas where an artistic hand is needed.
I suspect Generative AI is going to be something like that. It’s a great prototyping and idea-generation tool. Keep in mind that most of the examples we see are the good results that have gone through many iterations. I created something like 50 variations of the following scene before I got to this one.
I’ll certainly have more to say about various aspects of this technology in future newsletters. In the meantime, I encourage you to sign up for Firefly and try it yourself.
In the meantime, I want to know what you think. Is this the end of photography? The beginning of Skynet? Reply to this message or leave a comment. (Based on what I’ve seen so far, no one is using the Substack app and its Chat feature, so I won’t bother there.)
Insta360 Flow
I first became aware of the company Insta360 when I wrote about their early 360-degree camera for The Wirecutter (that article has since been updated and replaced by one covering action cameras). Since then, they’ve created what many photographers and videographers consider the best 360-degree camera, the Insta360 X3, and a novel webcam, the Insta360 Link.
Now they’ve just announced the Insta360 Flow, a gimbal for phones that uses a lot of impressive AI technology. I haven’t yet had a chance to use one, so I can’t comment on how well the features work, but as someone looking through the lens of computational photography, it’s really exciting.
The Insta360 Flow as a gimbal is impressive enough. It folds away compact and opens easily, with a magnetic mount for clipping your phone to it. (The phone still needs a mount; it’s the mount that connects to the gimbal arm that is the magnetic part, not a handy MagSafe direct connection.) From there, the gimbal keeps the phone level for smooth shooting or recording.
The hardware design is smart, too, with a built-in tripod stand and an extendable selfie stick arm.
But the software interests me the most. Using the Insta360 app as the camera app, you can lock onto subjects and keep them in focus or aligned in the frame using its Deep Track 3.0 technology, even when the subjects move or pass behind objects. It can follow you around, stitch panoramas, execute dolly moves, create timelapse and hyperlapse videos, and more.
To see it in action, check out this video by videographer Jeven Dovey:
So much of AI technology is geared toward making things easier for the operator, from autofocus to scene recognition to creating professional looking camera moves in videos. I can see how video creators would love this. And at $160, it seems like an affordable choice. I’ve requested a review unit from the company; I’ll report back if I get the chance to use it.
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.
SkyNet makes pretty images, with enough prompting. We appreciate your calm perspective of using generative AI to kickstart a creative process, rather than supplant it.