Structure reference in Adobe firefly generative ai enables you to copy from a reference image. Colin Smith shows you how to generate predictable ai from photo and sketch references.
► Free Photoshop add ons: https://photoshopcafe.com/vault
► I'm a BenQ Ambassador and use their monitors: https://www.benq.com/en-us/campaign/designer-expert/home/colin-smith.html
► THE GEAR I USE: https://www.bhphotovideo.com/c/browse/my-recommended-gear-colin-smith/ci/56280
► https://www.amazon.com/shop/photoshopcafe
► THE MUSIC I USE: http://share.epidemicsound.com/photoshopcafe
► SUBSCRIBE FOR MORE VIDS: https://www.youtube.com/user/photoshopcafe?sub_confirmation=1
► PREMIUM COURSES: https://photoshopCAFE.com/video
► INSTAGRAM: https://www.instagram.com/photoshopcafe/
► WEB: https://photoshopcafe.com
► FACEBOOK: https://www.facebook.com/photoshopCAFE
► TWITTER: https://twitter.com/photoshopCAFE
#AdobeFirefly #generativeai #StructureReference #CommunityxAdobe
AHey, it's Colin Smith here from photoshopcafe.
Com. Today, I want to show you a brand new feature inside of Adobe Firefly Image Generation, which
helps it to generate images based on reference. It's called Structural Reference. Let's have a
look at how it works. So go to firefly. Adobe. Com, sign in, and then we're going to go down
to text to image, and then we're going to hit Generate. So what we're going to do is let's grab
this, and why don't we do, Women Taking Self with phone. Okay, s
o based on this, we've got four
different photos. But what if I wanted against an isolated background and I have more of an idea
of what I want? Well, I can upload a structure reference. Watch this. We choose upload image.
We're going to grab this picture of a selfie here. And now we're going to use this to help us
create the same pose with an isolated background. So let's choose Generate. And if we look here,
now it's generating images in the same pose. Now we can increase the strength, wh
ich will make
it closer to the original. Let's hit Generate. If you notice, they're now holding flowers as well,
so they're now much closer to the original image. Now, notice it's not taking the color
or the style, it's just taking the structure or basically the outline. If
we want this on an isolated background, we could also type in and let's hit Generate
again. Here we now have an isolated background. If you're getting any value out of this video,
hit that subscribe button, turn on noti
fications, you won't miss any of my videos. And notice
it's getting much closer to what our original photo is. Now, if I take the strength all the
way down, it's going to give it more lenience to generate something that's not so close to the
original reference image. And as you can see here, there's a little bit more variance now. And by
the way, this is using the Firefly image 2 model, which is doing a lot better on humans
than it did before. Notice we have five fingers and faces look like
real faces now.
So let's tie a different approach with this, and let's see if we can create some concept art of
a spaceship. Okay, so we're going to tell them we want photorealistic concept art. We're going
to get rid of the reference. We're going to make sure we're doing a phototype, and And so
we're just basically starting with a new prompt. Let's do this now. All right, we got
some pretty cool-looking spaceships here, actually. This is really cool. All right, so I
want to make it resem
ble something like, say, a type of insect or something like So why don't
we have a look at using a reference image? Now, I've shown you can upload your own, but you
can also browse the gallery of what's here. So we can see there's different reference images
in here already. So why don't we do something like this bird? And let's see if we can create
It's going to be something that looks... It's going to be influenced a little bit by a bird.
So there's a structure reference showing here. Hit
Generate. And if you look, these are
now starting to resemble the bird. Okay, let's create I need something completely
different. Let's hit clear. And this time we're going to do a castle. And then choose generate.
All right. We've got some pretty cool-looking castles here. So what if I've got something in
mind, I'm going to upload one of my own sketches. So I'm going to upload a sketch of a castle.
And so this can tell it, hey, you know what? This is the angle I want and the scale. And the
n
we can hit generate, and now it I should produce something that looks closer to my illustration,
which we can use to help guide it. And you can see now these are much more resembling the sketch
that I created. Let's go full strength and see what we get. And if you compare these, this is
quite remarkable, really. It's even duplicating the towers and everything that I put on my sketch.
Why don't we try something even more interesting? Let's clear this out and let's go a little more
fantasy
with this. Let's go to... Let's do a steel robot with electric arms. All right, this is a
cool robot, but we want a full body. Why don't we do? We could type in full body, or we could
just go to a sketch that I've created. So let's go ahead and upload a sketch, upload image, and
here is a robot sketch. So we're uploading this now. Let's generate again with the structure.
Okay, so we're getting an illustrative of style here. So these are definitely matching the robot
design that I put in. S
o why don't we change this? We want to go for more of a photo, and we're
going to hit Generate again. I'm going to drop the strength down a little bit to about the
middle because I feel like it's trying too hard to copy the illustration and it's not giving
it enough room for things like the head. And then we look at images like this. If we want to
download them, just click the Download and that will download them. Okay, let's try modifying
the a little bit, and then we're going to hit Gener
ate one more time. Yeah, this is getting
closer. Maybe just change this to Menacing. This is how we're using the new structure reference
inside of Firefly. So this is going to make it easier to mix AI with our own graphs and our own
art, creating more predictable results. We can use them for things, for composites, for creating
elements and different things like that. Remember, the structure reference guides the shape
of the object, the size and positioning. If we want to do things like sty
le, we can
upload a different reference for the style. So this is a newer feature inside of Adobe
Firefly. I'd love to know you guys' comments. Let me know underneath. And if you're new to
the cafe, hit the subscribe button, turn on notifications. You won't miss any of my tutorials.
And until next time, I'll see you at the cafe.
Comments
Amazing! so exciting the future is!
Informative tutorial on some great technology!
I'm a photographer, not interested in generative stuff.
Great tutorial, it's a really useful addition thank you for the heads up👊
thumbs up again!
I really like the tutorials you guys present here but I’m so fed up with AI, as used in art. I’m so glad I’m a boomer on the way out. To me AI is like an invasive species to avoid.
SD reference controlnet has been available for over a year... We can also do poses (rigging), lineart, depth maps, and so much more. Just got new ipadapter nodes this week.
Can AI fake money?
Looks a lot like Controlnet. I think Adobe is going to be a grolillain this aspce. Taking a long time but I guess it's better to get it right. Censorship is still thier biggest issue in my option. I see the need to censor some things but I think they are being way too heavy handed. Drives a lot of people to open source.
Thanks!
This is an absolute game changer of a feature.
Great demo Colin. It's amazing how quickly the quality of generated imagery is improving.
Majic
As always, you have the best and easiest explanations, showing all of us how we can use these new tools to improve our art!
They're getting so close to where opensource SD was a year ago :o
So, each day is more closely to watch people who will only write commands and choose the images that an application suggest versus people that will work with cameras, lights, scenarios and more real stuff. I will always support the real ones.
Normally I don't share your enthusiasm for most new features but this one is good. Only criticism is that the background part of a photorealistic image lacks depth. I have noticed that the algorithm here seems much better at recognising what you want from the text input than generative fill in Photoshop
Let’s be real, Adobe firefly is the worst of all AI generators. Humans still look like straight out of horror movies and so on.
Wow, great improvements on the faces and limbs, which is one of the current issues with Gen Fill in Photoshop. I'm in the middle of a composite project right now for our class and I want to try this. Wahoo Adobe!
Very interesting...