Meet Firefly Image 3: Adobe's New AI Image Model

0:00The wait is over. Dive into Audible's most anticipated collection, The Best of 2025. featuring top audiobooks, podcasts, and originals across all genres. Our editors have carefully curated this year's must-listens from brilliant hidden gems to the buzziest new releases. Every title in this collection has earned its spot. This is your go-to for the absolute best in 2025 audio entertainment. Whether you love thrillers, romance, or nonfiction, your next favorite listen awaits. Discover why there's more to imagine when you listen at audible.com slash best of the year. Adobe has just announced a brand new AI model, and that is a new image generator.

0:47It is the Firefly Image 3. This thing has a bunch of amazing features. I'm going to be going over all of them, showing you some demos. I actually got my hands on using this product recently, and I've been really impressed with a lot of things that Adobe has done. So they did a big press release that kind of outlined everything going on here. But you know, the funny thing is where I get a lot of my news these days is literally just from X. So I actually saw this first on X. If you're interested in following like an influencer that is in Adobe, Chris Castellanova is an awesome influencer. She works at Adobe and she always is kind of tweeting about the latest things.

1:24Anyways, she shared a really interesting video recently of some images that this thing can do. And the main thing she said is that it's good at generating realistic images. So this kind of new version, and we saw a very similar kind of progress with like Mid Journey, where it started being really good at kind of fantasy, and then it's kind of good at kind of like different type styles of art. And eventually this thing is able to just straight up generate what looks very similar to like realistic photos. photo realistic photos. This is, I think one of the big things that they are focusing on. So people have been commenting on this.

1:59Um, different people have been like doing their own tests and showing, you know, what they were able to generate with it. Uh, people have, you know, examples of, uh, models with blue hair and blue dresses and whatever, but like they look photo realistic, like an actual person. So it's, um, it's very, very good. Um, someone asked an interesting question and I'm going to cover this because I didn't see this in any press releases or anywhere else. So this is, it feels like Twitter is kind of the only place you get this. Someone said, do you think it has character consistency? Chris said, there's style and structure consistency.

2:35There's a way you can trick it to some sort of character consistency. She has like a tutorial and video where she kind of showed you how to do that. That being said, this is not something that specifically does. So definitely Midjourney is still ahead of the curve on this because Midjourney recently unveiled a feature where you have character consistency, meaning that you could have, you know, the same character in different images in different places. So that is definitely something that Midjourney has ahead, which is kind of a big deal if you're trying to do, you know, some sort of like book or some sort of graphics on your website where you need like the same character in multiple different positions or places or shots.

3:20This is something that Midjourney is now able to do, does it fairly well. I think it kind of made this big announcement of the ability to do this back in March. So Midjourney has been doing this for a minute. It seems like Adobe is sort of catching up on a lot of regards, but it doesn't seem like this one is something they've specifically caught up with quite yet. So hopefully this is something they're going to be able to do. One thing that I wanted to say is there was an interesting blog about this whole thing on Tom's Guide and Ryan Morrison published an article on this. Just one thing I thought was interesting.

3:52He says, Adobe says the work on Firefly 3 was focused on speeding up ideation, allowing designers to go from an idea to a fully fledged image in as little time and with as little friction as possible. I think they've achieved it. Unlike MidJourney where you have to learn multiple parameters and how to implement them, Firefly has a series of well defined and clear menu options. Now this is something true. I've tested this out and I'll show you some demos, but they do have a lot more menu options, right? So think dropdowns where it's like, what style do you want? Is it ultra realistic? Right? So they kind of have those, which feels a little bit, you know, it feels like there's less learning how to be less hardcore, really good prompt engineer.

4:29One other thing he said was Firefly 3 seems to have better photo realism, a wider variety of outputs from a single prompt across styles like photo art illustration as well as options to set mood or lighting. So it's got a lot of really cool features. When I was playing with it, I actually asked it to generate something. I've done a couple things. The first one I asked it to do was to generate a picture of like a house and a cottage. It's kind of this default prompt and it did a great job. The one interesting thing I will say is I put the words hyper-realistic, which if you do that on Mid Journey, it was hyper-realistic.

5:06the image it originally generated for me was not and then again I kind of changed it and asked for like a castle I said a wooden castle and said for it to be hyper realistic said for there to be statues it actually generated a it looks like a photograph um is it perfect I would say probably not but like it actually looks pretty dang good there's a person in the background there's statues there's like mountains uh there's a there's a castle with like a bunch of wooden elements to the castle, which is what I asked for, which so I thought it was kind of cool. And of course, it does, you know, like a lot of a lot of these image generators, it generated four variations.

5:41And I will say, you know, some of them were better than others. Some of them, the windows were like filled in with bricks, and others looked a little bit more realistic. So yeah, you kind of get what you pay for. But I think as far as as far as this goes, similar to mid journey, you're going to be The ability to change the aspect ratio if you want it to generate squares or if you want it to generate portrait, widescreen, landscape. That's really nice. I don't see that on all of them. Content type, you can get it to do a photo or more an artistic style. And you can kind of toggle that on and off so you can see what you want it to do.

6:23So yeah, there's a bunch of really interesting things. I wanted to go over some of the very specific features that they've actually unveiled. So one of the big ones is called reference image. So with reference image, you're actually able to upload an image and get it to generate something similar based off of your prompt. So here's a quote from Ellie Greenfield, who's a chief technology officer for digital media over at Adobe. And Ellie said, prompting is a pain in the butt. Why spend an hour trying to craft a three paragraph prompt? if you have an image that you've created that's exactly the thing you want to reference.

7:01That saying a picture is worth a thousand words applies here, right? So you're actually going to be able to be like, I have this great picture of a castle. Now I want it to generate me a picture of a castle, but to change this, the element, or that element, you can literally use your own images instead of a prompt, which I think is amazing because I do this exact same thing with ChatGPT. For example, it's like I want it to write me a specific type of tweet or a specific paragraph of my article. I don't want to think of like how to like what my style of writing is to get it to help me clone that I just want to say like write it like me so what I always do is I just take an entire you know article and plug it into my prompt and I'm like hey write me a paragraph about x y and z do it in the style of in my previous you know writing style here's a excerpt that I've previously written it just reads that and then boom it'll output something that's similar to how I've written in the past.

7:51This is amazing. Wasn't really possible to do this with images. So Adobe is making it possible. Now there's a lot of people that are kind of concerned about it for a couple different reasons. So I want to read a paragraph from an article by The Verge that kind of discusses this. They say, users are expected to have the rights to use images they want to reference. Greenfield told The Verge that a message will flag this ownership requirement when the tool is first used and that the company is working on a universal do not train tag for Adobe's content authenticity initiative that it's also going to block images from being used as reference.

8:29So images uploaded as reference materials won't be used to train Firefly. Okay, first off, I think that's great. You know, it'd be super annoying if you're like uploading a picture to do something and then like everything upload is just getting sucked into their giant data set. I don't think people want that, especially corporations. So that is, I think, a good thing. They also said, quote, despite the ownership responsibility being placed on users, Adobe says this new referencing tool is still quote, safe for commercial use. One of the most notable advantages that Adobe claims Firefly has over rival gender of AI models.

9:01So they're saying this is still good for commercial use. Um, everything is trained with their dataset. Adobe has been really big on this. And so I think this is going to be interesting. All this is getting baked into Photoshop, um, which is again, really exciting. Uh, there have a couple other generative AI tools that are kind of already in here. And a lot of the stuff is stuff we've been seeing, right? They have like a generative AI generate background where you can like upload a picture of your product, for example, and then it can come up with a bunch of different backgrounds. You can say like, you know, they have an example where it's like a perfume bottle and then they're saying like generate a background with pink peonies or orchids or water.

9:37And it's like changing the background behind this perfume bottle. They have another one called generate similar. This is essentially going to be used where it can generate images similar or come up with different variations of the same image, which is really interesting. And then their third, this new model they have has a lot higher quality of capabilities in generating these images compared to all of the, you know, its previous ones. So it's going to be able to do a lot of interesting things. They have something called an adjustment brush that they've recently added. They showed a demo where like essentially, honestly, it reminds me of something that's already in, um, kind of like Adobe, um, Adobe shucks, I forget whatever the one, all the photographers use where they're creating the filters and stuff, uh, Lightroom, there you go.

10:27And essentially they have like a brush where they highlight someone's glasses and they can change the color of his glasses, um, with it. So, I mean, that feels kind of like a, an Adobe, uh, tool that you've seen other, seen other places, but this is something that's also gonna be baked into this. So all in all, I think this is really impressive. I mean, the big news here is just how much better this model is getting at generating realistic like photos. Not perfect. And I don't actually think it's as good as mid journey right now. It's kind of interesting when you hear people talk about this, they're saying like, this is so much better than mid journey because you don't have to learn these complex prompts.

10:58Like, yeah, but at the same time, the quality of mid journey, I think is still a little bit better. So like maybe it's a little bit more complex to learn the prompts, but if you want like the best in class, it's going to be mid journey. So all in all, I think that this is a really interesting time to be making these photos. I'll keep you up to date as they roll this out and add new features. But all in all, I think this is an amazing update. Great work from Adobe, really keeping up with this new image model here. And the thing that's amazing about, of course, Adobe that you're not going to see with other people, even like Midjourney, is they already have such popular software tools in the creative space and image and video generation.

11:39or, you know, editing tools, that now that they can plug this in, that's millions and millions of users that are going to get direct access to this. So I do think, you know, you can't discount Adobe. You can't count them out because they have such a huge user base. This is already plugged in and they just, you know, integrate this into a tool and millions of people will instantly be using it. If you enjoyed the episode today, I would really, really appreciate it if you could drop a review, like the video if you're watching on YouTube or follow us on Spotify and Apple, leave us a review. I love them all.

12:09Hope that you all have an amazing rest of your day.