Exploring Adobe's Firefly 2: A Deep Dive into AI Enhancements

0:00Adobe during its max conference dropped an exciting update to its generative AI image creation tool which is called Firefly image image2 model is now even more than ahd These features have traditionally posed challenges for AI driven models. Of course, we've seen, you know, people like Mid Journey and Dolly. I mean, I think Dolly might even still struggle. It's probably a little bit better, but like hands have notoriously been bad where, you know, when you generated an AI image, there was like six or seven fingers on the hand. There's just like a weird finger poking out all over the place. So, you know, it's definitely been something that's been a bit of a challenge and it appears that Adobe has finally kind of figured out the solution.

0:49And to be honest, you know, Midjourney has had this for a while, so they definitely get kudos as being the first player in the space. It's kind of funny because I know a lot of different artists that would use this for all sorts of different projects. And every time they would generate a person, they literally would have to go into Photoshop or some other image editor and remove extra fingers or extra random kind of things and make everything seem, you know, work good. So in any case, I'm sure all of them are, you know, giving a sigh of relief and excited that Adobe is kind of has this fixed. So in a statement to I think it's increasing popularity.

1:21Adobe share that users have created up to 3 billion. That's with a B images using Firefly since its inception just six months ago. so breaking this down further a billion of those images were created just last month which is really crazy i think interestingly 90 of these users were previously um you know unacquainted with adobe's suite of products meaning um most of the people creating these images were not like you know they didn't have photoshop they didn't have premiere they weren't people that are used to using all the adobe products so i think this is definitely awesome because this shows adobe that they're able to get new users with this kind of, with these new tools, they're able to really target a whole new audience, which was not paying them before.

2:05I think this surge in usage led Adobe to transition what began as simply, you know, Firefly demo site into like a full blown member of the Creator Cloud family, which of course is Adobe's, you know, grouping of software that they have. So Alexandru Koston, who is kind of the mastermind behind Adobe's generative AI and Sensei division, he essentially kind of talked a little bit about this and talked about the enormous advancements powering the new Firefly. So he mentioned that not only was the model schooled using, you know, I mean, essentially the model was trained using the latest images from Adobe stock and other commercial sources, but it's also grown in size almost 3x.

2:47So he said, quote, think of it as a brain that's three times larger, rendering more beautiful pixels, enhancing details for the user end quote so Koston kind of talked about this and Adobe has also kind of doubled the data set providing the model with an even richer understanding of users demands so though this expanded model requires more resources Adobe promises the same speed as the previous version which I think is impressive and Koston emphasizes that while they're investing in optimization techniques the focus remains on the quality of the output so you know if they were like, you know, what are we going to, what are we going to prioritize?

3:25They're not going to prioritize it doing it fast. You're going to prioritize it doing it well and making sure the outputs are really high quality, which I think at this point, um, is definitely important. No, one's doing this incredibly fast. Even mid journey has a bit of a wait time, but I think whoever can do it the best, like users, if they're actually going to use a picture, cause I know a lot of people that are using these for like graphics. I mean, I've seen people like go on mid journey and make coloring book pages and use it for actual books they're making, or, you know, graphics for a children's novel like all sorts of things I've seen people use this for and I don't think anyone in that case cares if it takes you know one minute instead of 30 seconds or two minutes instead of one minute when you're comparing this to you know what would traditionally take an artist like a week to draw some of these pictures so I think you know the order of magnitude and the speed is just insane how much faster this is able to generate images and I think people definitely would prefer Adobe focus on the quality.

4:19So users can currently access the model via the Firefly web app, but there's also plans in place to integrate it into, you know, Photoshop and all the other Adobe cloud products. So Koston kind of emphasized Adobe's vision for generative AI, saying, quote, our customers often enhance existing assets using generative capabilities. It's more generative editing than just text to image. So I think that, you know, the Firefly web app is also rolling out a bunch of new features and users can now adjust the kind of depth of field incorporate motion blur and also tweak field of view settings which is really interesting um one intriguing feature is firefly's ability to mimic the style of an uploaded image so you know you can if you're a photographer and you have kind of a specific style you can you can upload a bunch of images that you've taken and have it kind of mimic that so in addition i I think Adobe has also added an autocomplete feature for image prompts designed to kind of streamline the image creation process.

5:20So if you're, you know, creating an image, it's going to autocomplete the prompt, giving you ideas on how to actually make your prompt better and, you know, generate what you're trying to go for. So dive in a little bit deeper into Adobe's innovation pool, Illustrator, which is of course the vector graphics tool, is now jumping into the regenerative AI landscape. They announced at the Max conference that the Firefly vector model is essentially pioneering the creation of vector graphics using AI. So this model can create entire vector graphics from scratch, and it was trained using data from Adobe stock, meaning they own the licenses to all of the content that it was trained on.

5:58And if you, you know, create vectors from this, you're not going to get sued, which is incredibly, you know, awesome for a lot of enterprise customers that they're not getting that kind of protection from something like say mid-journey. So this whole upcoming beta version of Illustrator will essentially empower users to create complete scenes using just a text prompt and Koston elaborated on this milestone saying quote, the journey of training Firefly using millions of vector images was challenging but rewarding. Our aim was to produce images with minimal points. So Illustrator is often, you know, is also kind of unveiling a feature termed as mock-up and this is going to allow users to superimpose vector art onto 3d scenes and you know that could be like a design for a can or a logo on you know apparel for example Kostlin said quote mock-up offers a dynamic way to present designs in real world context right so if you've ever wanted to put your design on the cover of a fake book or put your design on you know essentially any 3d object this is going to be an awesome way to do that.

7:02So, you know, kind of rounding off the list of new features is a new one called Retype. This tool converts static text in images into editable text and even identifies matching fonts. And this is something that I absolutely love. In addition, Illustrator is now web compatible, so you're not just going to have to download the app and use it on your computer. This is another thing I've been calling for for a long time. It seems very archaic. Feels like companies like canva are really really beating um all the adobe suites uh when it comes to like just being able to edit something in your browser figma obviously was really big in this regard and i think that's why you know adobe had to buy it because it you know adobe xd just wasn't keeping up um i think this kind of suite of innovations signifies adobe's commitment to really pushing the boundaries of what's possible in the realm of image creation editing this entire creative field um and you know I think it's making it really exciting to be in this whole space of creatives and all that kind of stuff so very excited for all these new features and I'll keep you posted on how they work as I test them in the future