Train your own AI model in FLORA

Lightly edited from the recording for readability. Wording is the speakers’ own.

Hello, hello, everyone. Welcome back to, oh, there it is, camera. Hello, welcome everyone to back to another FLORA live stream. I'm your host, Sasha Zabegalin, a forward deployed creative here at FLORA. Really excited to have you all today. We'll be covering LoRAs in FLORA, how to train image models in your own art direction, in your own visual aesthetics. There's a, as you can see, I have the canvas open already.

We have a lot of various examples, a lot to cover today, but we'll dive through them slowly but surely. I want to start off and say that this canvas was genuinely only possible thanks to our incredible feature FLORA MCP. What's up, Lloyd? Welcome back. Good to see you.

This incredible canvas that we have today to show is only was only possible thanks to specifically our FLORA MCP feature. Very wild. And I'll dive into how I specifically created this huge example workflow that covers a ton of different viewpoints on specifically inside of FLORA. But yeah, so before we start getting into LoRAs, I kind of want to blow your mind a bit.

So let's say we're looking at the canvas right now, right? If I click Exit Fullscreen, you can see I'm currently actually in my ChatGPT codex agent, having it continuously build more of these examples. So how I started on today's workflow was specifically I asked my agent to help build a huge example of LoRAs in FLORA, because there is a few models. We have the Recraft Styles, and then Recraft Styles Vector. We have Flux 2. We have Krea 2 LoRAs. We have a bunch of models that have LoRAs capabilities inside of them.

But inside of those LoRAs capabilities, as we'll dive into just a little bit, there is a ton of little things and tweaks and features. And I wanted to make sure today we covered all of them. Traditionally, this would take me hours of spending a lot of time on the canvas and building out all these workflows and like really trying to think through every single prompt in every single idea and possible direction. And I wouldn't be able to cover all of them.

But thanks to FLORA, MCP, I can just ask my agent to take care of that. So I started off the workflow by just chat, opening up codex on my computer. And I said, Can you build me a FLORA workflow that covers all the use cases of LoRAs in FLORA, specifically, create LoRAs for every single setting there is, and every single model that can take in LoRAs. Specifically, there is Flux 2 to create LoRAs as well, Recraft Styles, and there's a ton of different things for Recraft Styles. I want to make sure you cover all of them. Can you find various assets either by pulling from the internet, make sure they're copyright safe, or using the FLORA brand skill to create that visual, the visuals first to train LoRAs. Make sure the style varies among things.

Also make sure you cover that in the prep. There are specifically LoRAs for persons that don't have to check an object. I want you to create assets for those and that makes sense and cover various different art directions that someone might want to use LoRAs FLORA. Keep this workflow organized, clean, legible, and section off. Once you create those LoRAs, create example outputs across them that might be useful for people to visually learn. Create a ton of examples across all of those. Again, make sure you use FLORA and make no mistakes.

I told my agent to do this. And there was a lot of texting forth. We'll get into that in just a little bit. But it created all the examples that we have here today. I even ended up asking we just added our feature called decks in FLORA, where you can create power decks inside of FLORA natively. And so I asked my agent, let's pull up my agent chat. I asked my agent down here, somewhere.

There's a lot of threads. I asked my agent to create a slide deck. Where are you? Let me keep going. Sorry, my computer is really laggy because I have a crazy canvas and I'm trying to live stream at the same time.

So a little difficult. I hope it's not lagging. Okay, so then I asked specifically my agent. There's a lot of me going back and forth with the agent. It kept spawning all these annotation boxes and I was trying to make sure it wasn't too big.

But anyways, I said for the slide deck, could you possibly include FLORA UI screenshots to showcase as best as possible? Thread before, but I just told chat community. Hey, can you make, can you use FLORA and make a slide deck using some of the assets we've created today about LoRAs and FLORA to high level walkthrough how FLORA, LoRAs inside of FLORA work. And it did. I will also share this canvas with you all as well. I'll put it in the chat. I will set the view link so you all can I put the Canvas link in the chat, specifically on YouTube, if you want to check it out there.

I'll also publish this Canvas in the community page later today once my agent is finished working on it. So there might be some missing pieces if you join in now. But here is the deck that my agent created. So a familiar look, a new possibility, LoRAs and styles inside of FLORA. We'll learn from real examples, what to train, which models to choose, and how strength can change the results. And here is an example output of specifically applying this tactile paradox LoRA recreated inside of Flux 2 with the strength set at 2.5.

So LoRAs teach a model a specific art direction and a specific vibe. And then you can ask once you create a LoRA, you can prompt anything really, and it will apply that specific look and style to it. So for example, we could train a LoRA on a bunch of these types of images, this kind of abstract, tactile vibe. And then I can type in a prompt like cat, and I will get an image of a cat in this sort of art direction.

So LoRA adds a learned look or subject to a compatible image model. You use imagery for training LoRAs. And what you want to be very much sure of is that the sets of imagery all kind of come from the same art direction and creative world. That way, it's like the most consistent as possible.

So for example, we will change a bunch of these images. And then now we're able to prompt a new subject, a floor lamp, and we'll get a floor lamp in that sort of paradoxical style. For deciding what should stay familiar, the style, person, object, and character training help keep a different part of an image familiar. A style is something like a specific art direction. You can train a LoRA on a visual vibe. You can train a LoRA specifically on a person to help maintain that consistency and identity. You can train a LoRA on an object.

So then you can continuously get set of those products in those imagery. You can also train it on a character. So variety definitely helps with these LoRAs. But you also want to make sure you train with clean individual images, especially with rights you can use. Do not try to train LoRAs on other people's art. Use your own or your past AI work. That's what I've done a lot.

Particularly, I have a huge portfolio of pieces and I'm able to train a LoRA on those. So it makes it even easier to prompt into something like a model that's capable of LoRAs, like Flux 2, Krea2, and Regraft. You want to avoid things like frames or interface screenshots or accidental texts that you don't mean to include in this training. You want to relatively keep the look across those sets of images consistent across the board. That way you can really nail that look and feel and art direction that you're trying to achieve. Try to vary between each image. You don't want to train a LoRA on the same image over and over again.

It's not going to really do much for you. But if you have the same sort of art direction and just various different ways it's interpreted, it can help provide the model more perspective on that kind of universe. First, you want to source your imagery to then start creating new imagery. There is, again, different types of LoRA-capable models inside of FLORA. Each of them, unfortunately, have their own training.

So you have to train a model individually. You can't just train it universally across Flux, Krea, or Regraft. It's something to note. The Regraft style model is the fastest one. It's really quick. I'll show you in a second.

But it's up to 10 images. You select those and it's like a blip. You get a style back. While Flux 2 specifically can take a very long time, which is why my agent is still currently working because I'm waiting for all the flux to generation LoRAs to be trained and up on this canvas.

But Kraya is not too bad at specifically lower trading. It still takes a bit, it's like 10 minutes. Yeah, so each Lora is only compatible with that particular model, which is something to note. And then we'll dive into specifically training as well. You want to train it and then just give it some time. If you're using Kraya or flux, you have to just give it some time. You want to make sure you give your Lora a very clear name.

So when you go pick your Lora, what you're doing. And then choose a style and then start describing your scene, your vibe, whatever you're trying to create. And then see what happens. And then from there, you can compare a few settings. Each Lora model has different capable settings, like specifically strength.

So you can tweak the strength and we'll get into the examples specifically how what that does and how that affects it. And just a little bit, but they do massively change the strength setting does massively change how much it follows the Lora. And here is a screenshot of specifically the strength influence. So again, we have this tactile paradox or we're using we're using on flux too. We have the same prompt and seed number even. And these numbers are just specific for this comparison.

So again, it's the same prompt and the same seed using the same Lora. And we can see here when we set the strength to four, it creates something that it's completely hard to read. And it's not really in our vein of art direction there while strength setting of something like 2.4. 5 specifically for flux. It's a lot more steered and clear.

But when I set the strength to one, I mean, I guess it's in the art direction vein, but it's definitely losing a lot of its shine. But then if I set the strength to zero, it the Lora is turned off entirely and we don't get at all what we're there. So strength is definitely helps influence your generations and guide the model towards that desire. And so. Oh, wow.

OK, the agent is adding stuff to this. One second. Let me go back there. See a comment. I have tried to train Lora on for a detail object using flux for just couldn't get it to keep the object consistent. Is this something you're going to show? Yes, we are going to take a look at that specifically.

But I personally don't from the examples that we have didn't do too bad with objects and people for Lora training, but I don't recommend Lora's for specifically style training. I recommend specifically character object training. I recommend Lora's for art direction because models always tend to struggle with art direction like transferring while. But like banana and see Dream and GPT do really well with like character and object consistency.

So I usually use I only really use Lora for making imagery in my same creative world. But OK, let's take a back look at the slides. Again, I didn't make these slides. This was our this was church B6 controlling FLORA using our FLORA MCP and building out this entire deck and workflow that we have today. Really crazy stuff, but it's possible now with FLORA MCP and how far you can push it.

But let's keep going through. Let me see what. Let me see what it's done. OK, great.

OK, great. The agent put in screenshots. So to get started with training in FLORA, you can you just start off with uploading your pictures, you can show it as well. Let me.

This is a PDF. And let me back to the canvas. Let me put that there. I'm going to extract this PDF for the canvas so I can just see it all at once. I have to do one second.

Okay, while that's loading, we can just click on the image node itself. Yes, Lloyd, if you upload it, if you connect your power, if you drop your power, like if I drop an old PowerPoint, if I drop like an old PowerPoint into my agent while I'm using FLORA MCP, I'm hey, recreate this PowerPoint deck in FLORA. I'm using FLORA's deck feature. The agent will take care of that.

But make sure you're using codecs and potentially attach the fonts as well that you're using. That way you can upload the fonts directly to the canvas as well. But yeah, okay. And so let's go. Recraft. I'm going to specifically show our training first with recraft styles, because it is the fastest one to start from scratch.

Let's go recraft styles. Okay, we have recraft. Okay, recraft styles. So we have recraft styles as a LoRa model. We have flux to LoRa. We have craya to LoRa. And so any of those, there's a, I think there's two more actually.

Let me duplicate that. Okay, why would she duplicate? Okay, there you go. Recraft. There's also recraft styles pro vector.

So it's recrafts vector image model. So you can train a specific vector style, which is interesting. We'll get into that in just a little bit, but any sort of a LoRa capable model, you will see in the right hand side right here, you'll see this LoRa feature right there. In this case for recraft calls it styles, but flux to you and craya call it LoRa. You'll see this toggle right there. And you can start creating one or pulling from your library directly. When you click create, specifically with recraft styles, what you have to do is just upload a set of up to 10 imagery.

Let me open up my computer right now and find a folder that I have. Okay. Let me upload those. So you just drag and drop in your imagery. Recraft specifically, LoRa training is the easiest one to get started, which is why I'm going to show this one first, because it takes a lot less headache. You just upload your imagery.

And then you have this toggle right here called style match, where then you can prescribe which kind of how you want it to affect it. So what style match does is how closely the generation follows the style by default. If you choose precise, it reproduces it's flexible enough to keep the mood and lets the prompt lead. But each generation can override it. Where you can go to flexible where it can be a lot more flexible in the visual sense of how it looks.

And then recraft styles can only take up to 10 images for LoRa training. But then we just call it I'll call it Sasha style. Let's just do that. Then you just create the LoRa and let it do its thing. I'm going to check in on my agent, seeing what it's doing.

Okay, it's still finishing up stuff. I'll give it a second. I think my internet's being whack. Oh, no.

Okay, I'm just going to refresh. I'm just going to refresh that page one second. But once the style is done training, Recraft is really fast. Your style will be done right then and there.

So you can start using it. And so let's-- where am I? OK, there we go. So once you do that with Recraft, all you have to do then is go to Styles here. And I think mine is called-- have mine somewhere.

OK, there it is. You can just select that style. You'll see all of your styles are here. Cool part is FloraMCP can build styles for you. It can train mores for you. And it can intelligently choose all the settings, which is honestly really great because there is actually a lot of toggles and things. And it's kind of always hard to figure out what specifically is the proper setting.

Specifically for Flux 2 and Krea. Flux is a little more scarier one. But Recraft's really simple. You just upload your things and choose if you want precise or flexible.

But then, OK, so now I have my style applied there on the right hand side. And I can just have my Recraft styles model. And I can say something like CyberCat on a white background. Apply that specific style. And let it do its thing. All that's doing its thing. There we have our FLORA applied.

We got our CyberCat trained in my Adidas fashion stuff. Yeah, it's really simple to get started with creating mores right there. Especially with Recraft. We'll dive into the other ones in just a second.

But I see a comment. LoRAs can take hours on a single GPU. Yes. But in our case, they run on the cloud.

But specifically, Flux 2 is the one that takes a really long time. But Craya2 and Recraft are relatively fast with Recraft being the fastest one. Why can't I extract? Oh, let me browse through the-- look, it's trained in strength. Flux. Cool.

So let's take a look at some of the other models. So also something to point out. Is you can access your LoRAs inside of your Asset Library. So if you go to Assets right here, and then you go to Styles, you will be able to see all of your styles. And so again, great part is you can use FLORA MCP to help create LoRAs for you. And I have a bunch of them. And you can see it's all with the different models.

Here's the same LoRAs. But for Recraft for Vector, here's a LoRAs. The same LoRAs, but it's for Craya2. The same LoRAs, but it's for Recraft v4, et cetera. There is a ton of options I've created for today to experiment with. And we can just browse through and utilize any of them in one second.

So let's say you have a bunch of images on the canvas. Let's say this is your art. You just input it onto the canvas. Again, you can just download any of these, particularly, and then open up the LoRAs training art right here.

Let's look at specifically the Krea model right here. Again, you go to the right-hand side right here. And then you go to-- you can also upload your own, specifically for Flux 2 and the Krea model. You can upload your own Tensor files if you have those. I know not everyone does, but if you do, it is possible to upload your own Tensor files.

But when you go to-- if you don't have Alora, you can go to Train. We can just upload our visuals in directly. Let me just save some of these. Let's say we have them.

Let's roll with this one. I'm just gonna save all of these, zipping up the assets. Let me open it up on my computer. Okay, go to, let's train Alora now.

And then we wanna drag in our visuals. Let me select all those, open, okay. So the create a two, Alora training, a little different. With the create a two model, it has to create captions for each image and describing it and stuff.

But the cool part is we have specifically, we have auto. So you can, it will automatically decide which labels to give to each image. But this is where you can choose specifically if you want to train this Alora as a style, specific aesthetic, creative direction, vibe, movement, or if you're trying to train this as a subject or a character specifically. So you can choose between the or.

So let's say you have a huge set of like art direction stuff, visuals like this. You would choose style for this case. But let's say you have a bunch of images of yourself or your model or a character you've built. You would choose something like a subject or character and then you would train that Alora on that particular thing. For the create a two training, you want to give it a specific style name. Call it tack tile.

And then for the create a two model and the flux to LoRAs, you have to give it a trigger word. And a trigger word, it means a rare keyword that carries that style. It's telling, it's like a prompt injection that tells the image model to use this particular as its guiding force. And you want to make sure your trigger word is unique. It's usually like a mumbo jumbo mess. Try to make sure your trigger word isn't like something that can be interpreted as a prompt. You wouldn't want to say like you would probably call this like blob style.

I don't know. I would call this like I would just call it gibberish. I would call it something like blobby XYZ, something like that, just to make it make sure it's not using like a real word because then it can diffuse the model and you won't get your style applied. And then there's also steps involved. If you leave it blank, it uses the recommended default automatically. I left my steps blank. What steps do is how long it will the LoRA will be training for.

So how many times it's going to run through the training process over and over again. The more steps you give it, this isn't the style, but it can overcook it can make it too much. So it accepts from 50 to 3000. I just leave it blank. I just go with the default. I don't I don't try to think too hard about that one. And if anything, I will ask my agent to help look online.

What does it think is the best recommended steps for my source material. But the cool part is, again, your agent can create a LoRA in Florida. So it will automatically fill all fill out all these details for you without you having to worry about it too much. And then this is also where you'll see on the right hand side, the strength influence where you can turn it up, turn it down. Strength influences how much that specific look and feel applies to that visual.

So again, something to note there. And then for the lower training process of flux to a name thing, you just choose again right here, LoRA, and you can again upload a safe dot tensor file if you have one for flux or you can start training your own by just dragging in your imagery. I'll just drag in my imagery again. And then this is where the flux to LoRAs become painful because unfortunately they do not have an auto label mechanism. You're going to have to label each input yourself with flux to but again, your MCP agent with FLORA can take care of us.

So highly recommend using the FLORA MCP to help you create LoRAs because it will save you so much time as you can see with my canvas. Like I just commanded my agent and it took care of all of that. But same sort of thing. You have to label everything. You want to give it a good style name. You want to make sure you give it a keyword, trigger keyword that is something very random, very strange. That way it doesn't trigger something with the prompt and like miss, misconstrue it.

And again, the steps usually I recommend just leaving it blank. That way it just sets it at the default level it thinks is the best amount for that particular set of imagery. Let's get into some examples. I do see a comment. Steven said, I'm into branding. How can you use the lawyers to train or maintain a graphic visual system in different applications? Can you please showcase any examples of how I can approach this kind of projects?

Yes. So this whole project here today, I will share also the workflow link one more time so you can clone it yourself. But feel free to follow along. I recommend specifically, LoRAs are best for art direction. And if you really have an established art direction and want to continue building visuals in that art directed world, utilize LoRAs because they're really great at that's what they're really meant for.

So we have a bunch of different examples. So on the left, how to read everything is on you'll see the title source pictures, meaning this is what I'm using to train a LoRA. So on this left hand side, we're using this tactile paradox right here to train this sort of LoRA on it. And then you'll also then see each you'll see each generation. It's the same prompt with different specific specifically the strengths all tweaked. And you can see kind of how each strength affects the output. And we do that across many different styles and many different things.

From here, we have this sort of graphical art style. And then we're trying to use LoRAs and seeing how it goes. And you'll also see it specifically across various different models. You'll see it with flux, you'll see it with recraft, you'll see with crayon.

So you can get a sense of how each one approaches LoRAs. And then here is here we created a huge data set of a single person. And we're going to train a LoRA on a person. And surprisingly, I think actually didn't do too bad. Back in the day before there was nano banana, and chat image to you had to train LoRAs on yourself on a bunch of pictures of yourself to create those sort of selfies of people or yourself using AI.

So kind of this throws me back to those early days before nano banana was really good. And you could do this. But you can see I trained my agent trained a style or trained a LoRA on a person. And you can see how it's being used across the board. Specifically, again, this is all just using the Crayo model here. And you'll see all the different strength settings. And you'll see the prompt as well.

And you can see it's using the keyword that trigger keyword in the prompts themselves as well. And I see a comment. I use flux to train my own avatar for founder short form videos. So can you actually do this with yourself? Unfortunately, I couldn't get in.

So I'm not sure if I missed re explain LoRA. Yes, you can use. If you have your tensor flux file, you can upload it into the LoRAs. And again, it's on this right hand side. Any model that has floor lower capabilities created to flux to recraft whenever you spawn it on the canvas and you go to the right hand side, you can select LoRA. And this is where you can upload your tensor file or train or pull from your library of specific LoRAs. Let me look at some cool examples because.

Yes, cool. Which ones do I want to take a look at? I like. Let's roll this one.

This looks cool. So let's say here is a bunch of here is our portfolio of pieces. We want to train a on this specific style that we think of all this sort of prismatic light, this kind of orange wash. We have all these images, we download, we have them on our computer, we download them, and then we train Elora and FLORA with these images. Once we have that Elora, let's actually just take a look at how it appears across many different prompts and different strength settings. Across various models from Krea to Flux to even Recraft and see, let's take a look at who do we think is the best at interpreting our visual direction. I think for the most part, I've already taken a look, but I think Krea 2 Lors are my favorite so far and we'll talk a little bit more about that in just a second as well in some interesting ways I've used FLORA MCP and our Chachapati 6 as an agent to go do really awesome things.

But before that, so here we have Elora applied, that prismatic light, Elora, somewhere in here, I don't know, there it is. We have that prismatic light effect. These are our visuals that we train Elora on. Again, this sort of grainy kind of orange, shiny look, and we train Elora. We waited, it was done, it was ready to use.

Now let's test it across various different prompts with various different strength settings. So in this case, this top row right here, we're using the same prompt of an unbranded class fragrance bottle on a round plinth with folded paper beside it. Simple product campaign, no composition, no text, no lettering, no logos. And then we made sure to, there is our trigger word right there in the beginning, just to force that style.

And then we'll see all the different various strength settings. So this is that same prompt, but at strength zero, this is that same prompt at strength 0.5. This is the same prompt at strength one in Kreia. This is the same prompt with strength at two, same prompt at strength four. And you can see the strength levels really do affect it. Overall, you can even see, especially with this row right here, you can see when we started going up to strength levels one and two, our LoRA really applied very well. And like it very much feels like it's part of that same like art directed world.

But you can kind of see with four, it kind of goes a little crazy. It kind of goes off the beaten path too much where it's like no longer has like a weird grain or it's it's called overcooked. It becomes a little bit overcooked if you set the strength too high. It's really fun to play with because a lot of the times I like mix and match the strengths. Sometimes I just use it. I have a general sense of a style and vibe. And sometimes it turned down that strength to like see maybe like a sister image to that one, because like visually, both of these are really cool.

This one definitely followed my prismatic light style a lot more. But this one was still also really cool. And all I did was just turn down the strength specifically. And then you can see when you turn it off very different vibe.

And then when you set the strength of a LoRA with Kreia to four, it gets real crazy. It's not even an image anymore. It's just noise. Then the next example is taking a look at that same prismatic LoRA that we made, but on the flux model particularly. And again, same prompt and unbranded class fragrance bottle on a round plinth with folded paper beside it. Simple product campaign composition, no text, no lettering, no logos. And again, you'll see the strength at the top right here.

I've laid it out with zero. It looks like a regular Schmegular product image. We turn it up to 0.5 with flux and flux still doesn't really do anything. We set the strength to one now with flux and still not there.

But then we set the strength to 2.5 with flux again, same prompt. And look, we were able look how cool we were able to do that. So then if we set the strength of flux to four again, it gets a little crazy and wonky there what's something really cool with specifically the flux model and I just learned this today I had no idea turns out you can plug in a reference photo that you then can apply that LoRA onto I had no idea the flux to LoRA model was capable of that which is really cool so let's say you have a product let's say it's this I guess vase I think this was supposed to be a vase or a maybe this is a Christmas ornament and you're a brand trying to create this new brand campaign for your vase or your Christmas ornament and you have a particular aesthetic art direction you're trying to go into you can train a LoRA on the art direction and then plug it in specifically with using the flux to LoRA model you can plug in that product and then use a prompt something like keep the supplied orbit vessel silhouette and proportions restyle its surface lighting and background in the learn visual style preserve the object as a central subject no text no layering no logos and flux did a really great job at applying that specific art direction to that object right there I mean it did lose a little bit of consistency but we were able to kind of spin a version of our object in a new art direction I see some comments oh yeah asks Sasha suppose the goal is to train a LoRA specifically for composition and lighting for instance a photographic style character is by specific cropping and color saturation will this work for me I would say probably not at some it's more so high level style not necessarily like keep this exact angle of things LoRAs are much more like as you can see I'm training a LoRA on this sort of glistening blue and orange glassy kind of vintage like vibes so that I can create more images set in this world so LoRAs are again very important for art direction and being creating the ability to create more visuals in that same visual world with relative ease because now I can just show you let's call it up image let's use flux to LoRA let's do I wasn't a prismatic light right there I just selected that now I can just say something like we have to remember what the key word was I wait they have different keywords sorry no wrong one here it is now I can do something like I need to make sure I my keyword and MJ flux prism cat sitting on a desk something like that oh wait and then the strength is a pretty good amount and just let it roll and then you can affect the quality I will set the quality at 100 let's get a really nice image from flux actually don't use the flux to model too much so a little bit new to me but it seems like it's actually pretty well there we have our cat sitting on a desk and it's coming out in this sort of same visual world so that's really where it comes in handy is it's I think LoRAs are the most useful for again art direction and world building and being able to create visuals and continuously in the same sort of style it's a lot easier and faster than using like something like nano banana or GPT Image or those various models because those I feel like can drift in style versus you train a style once and then I can literally just say cat on a desk right I don't need to say cat on a desk with a shining glistening glitter effect the model already knows that kind of what I'm what I want so for example my signature thing is like always a kind of blue futuristic images so I can train a model on a kind of visual sense of that let me find the cool ones I have some more cool ones like for example here I have this like really crazy kind of Pixar like colorful 3d clay like visuals and I can train a model on that sense of style you can kind of see how I'm able to then create more visuals set in that world so it really helps kind of world build continuously what's really cool is like here is another style of like kind of illustrated imagery and you can train a LoRA vector model with recraft vector yeah so you can train a specific with a recraft it's called recraft styles pro vector these model names are getting crazy you can train if you have a specific like illustrated style you can train a LoRA or a style with recraft on that and then create more vector pieces inside of that world because the cool part is when I find a recraft vector model this will be you can download this as an SVG file and so this will be an SVG file that you can mess with and I believe we've added actually vector support to the canvas or maybe that oh yeah we did cool I don't know if that was I didn't know if that was in yet we have actually let me make sure I put this 16 by 9 okay there it is spot a new one actually go there sorry my computer is gonna be kind of laggy because trying to stream and I have a crazy canvas open let me go there but the recraft models especially particularly they generate as if you choose the recraft vector models it will generate as a vector image so you can train a LoRA on your like vector look and aesthetic and then continuously produce like vector graphics that style and then still edit them as vector SVG files that you can take into all of your graphics and things like that let me delete that let's take a look at let's keep going here okay I see some comments so someone asked if you were a home goods brand and wanted to set your products in the mid-century modern Scandinavian mid-century modern Scandinavian style home would you train a LoRA with images of rooms and put your products in the images as well no I would just use nano banana or chat gbt image and just supply those reference images and then just ask it to visualize it in mid-century again LoRAs art it's more so art direction it's not the we will see that there is it's more so art direction product is fine like here we have an example of here let's say we have a ton of images of our product in different angles and shots and ways it's it's it's seen we can train a LoRA on that product particularly and then use it throughout you can see it doesn't do too bad of a job but I still I think it's I honestly think training a LoRA on a product is quite overkill in most cases because no banana chat gbt image and see dream I've gotten all really good at just carrying those that consistency over it's very small cases where to actually train a product as a LoRA if it's if it's like if the AI just really struggles with it but even then I think I still think it's just overkill because it's not gonna perfect it just because you train a LoRA on your product doesn't mean instantly it's every generation of it is gonna be exactly perfect and it's gonna result in that so usually stick I usually use only really use LoRAs for kind of style and art direction I guess people is good I just found that out I haven't used LoRAs with characters or people in a very long time actually since again nano banana and chat gbt came out because they've just been really good at transferring them over but it looks like it's actually not bad at that, so we trained, let's show a character, let's show a person actually, the same person. Yeah, we trained Elora on this dude, but let's say it's you, let's say it's a character you've created, you can train Elora, specifically Kreia or Flux, no, Kreia is the only one that can do the people and objects, LoRAs, but you can train Kreia on that person, and then it makes it even easier 'cause then you don't have to attach a reference photo of that person. I usually have only really done person-Lora training on either myself, 'cause I'm trying to put myself in crazy fantastical images, or if I'm working with a client where they want to campaign photo shoot, it makes it really easy, 'cause all you have to do then is just, let's say, in this case, you're saying the keyword, are we person hiking on a rocky coastal path, full body environment portrait, and we're able to create visuals there.

But it's still think, I don't do this too often, I'm mostly using LoRAs for art direction, and kind of creating more visuals in this world, it's really great for brand work, 'cause you can, what you can do is, you can create a mood board of images of your brand, you can create a mood board of the visuals, like let's say this is a new, I'm a brand designer at an agency, and I'm tasked with creating the visual world for a new health tech outdoors company, and they need a bunch of visuals, and they provided a bunch of their graphics from the past or something, or you've already made some, you've made maybe a few, you can then source them all together, and train a LoRA, and that way you can start creating more visuals that kind of follow in that same creative direction, so then you can kind of, they can even share the LoRA with them, so that they can use it themselves, so they can continuously create visuals set in that world, so they can, you can prompt, you can train a LoRA on this specific style, and let's say, you give them this LoRA, and let's say, they're oh, we need a new visual for cool background animation visual, for our new launch, right, and they can type in something like flower blossoming, and then the flower that the image is gonna generate, will visually look like one of these. Yeah, I see some comments. The finished LoRA saves in the interface, and they can be used in any prompt, or only in where they're created, so when, styles will exist in your assets library, if you go up here in assets, you go to styles right here, and you'll have all of them. You have to train, the LoRAs aren't universal, it's, there's only three models that are capable, technically four models that take in LoRAs, which is recraft four, and then recraft V4 vector, create two, and then flux two, but you have to train them all individually, but they will be saved into your library, and you should also get a notification, inside of the FLORA app, when a LoRA is done training, also I believe an email, when your LoRA is done training, so you'll know as soon as it's done, recraft is, it takes, once you're done, once you click run, it's already done, so recraft is the fastest style slash LoRA to train, while something like flux, two will actually take like a few hours, I can't lie, so just bear with that, but yeah recraft is the fastest one, and create two is in the middle of it all, and the training vibes. I believe we also have a case study about LoRAs and FLORA, where one of our end team, Alex, talked about why we brought it back, why we deprecated, we had styles, and then we deprecated it, and then we brought it back, but now it's more focused on particularly these four models, because these four models, or three models, accept LoRAs input, and it can actually use the LoRA itself and its visuals. What's really cool is it becomes like having mid-journey in FLORA, and you can import your mid-journey images into FLORA, and then train a style on that, and so you can use a model like create two, on those visuals, 'cause I feel like Kreia 2 is the closest we have to mid-running inside of FLORA, 'cause it has that very aesthetic visual sense.

But Flux is also, I'd say, on par with Kreia. As you can see, look at that. We have that cobalt chrome LoRa that we trained right there. We set the strength there up to four. I mean, the product completely went away, but we got this really cool abstract visual. It looked kind of like a seed or something.

So there's another question. Would you recommend for LoRAs trained on a person for the reference images to have the same lighting background style in our wardrobe? Would that be most successful for consistent outputs? It depends. I think training a LoRa on a person, it's all about actually having them in different outfits and visuals and different lightings, because that way the LoRa has more variety to think of when it tries to put your person into a new scene or new outfit.

But again, I wouldn't recommend doing a LoRa with a person unless you absolutely have to, 'cause the Nanobanana and Chatchity Image 2 and Sea Dream can do really well with just, here is your image of a person. Can you use Attacha as a reference and just say, place this person in X, Y, and Z background, and you should mostly be fine. The only times I ever used people LoRas was, again, way back when, before Nanobanana and Chatchity Image existed, 'cause back then there was not really a way to get consistent characters or people into image models without a LoRa. But now we've kind of, LoRas have progressed, image models themselves have progressed so far that it's not really needed, and it's also hard to describe and pinpoint when you do it.

So again, art directions where LoRas are really, really shine. But we can take a look at how Recraft interpreted that style that we created. And you can see the, it's different. The different settings are there of style match.

So at the top, we have it at style default. So it decides how closely to follow the LoRa itself or the style itself. And it's the same prompt. We're just using the same prompt. And, oh, wow.

So there's, turns out there's, there's Recraft V4 Styles. There's Recraft V4 Styles Pro. And then there's Recraft Styles Vector. So there's three different Recrafts. Recraft Styles Models for you to pick. It's a lot, I can imagine. The main difference between Recraft V4 Styles and then Recraft V4 Styles Pro is the Pro model is at 2K resolution.

So we'll generate the image in 2K quality while the non-Pro version is only at 1K. I mostly, I'm just using the Styles Pro. That way I just get a really nice crispy image and I don't have to later down the line, actually go ahead and upscale it again and spend more time and more credits. And also sometimes I feel like when I upscale, I feel like sometimes I lose the initial feeling of the image.

So I always just try to generate it with one shot with the resolution already being high. So I use those, the higher quality models right away. Both of them also kind of interpret it differently. I feel like the Pro version goes for a much more premium, hyper beautiful look while the regular Recraft model, not as much. You can see also the Recraft Vector model.

So we'll create vector images. In this case isn't the best use of this one because, but unless you want an illustrated, bottle in the style could be useful for you. We can see, oh, then we can also see when we set for Recraft, we set the style, we can set the style match of either precise or flexible. Precise means again, that it follows that style precisely and tries to match as close as possible. While flexible means it can interpret it a little bit more. It can be flexible with how it takes in that style. You can kind of see down here is the match flexible.

And I feel like it's a little too flexible as in I mean, it's not as cool and artsy as the original images. So precise in this case, precise in style default. I feel like did the better job. I see a question in the comment. Lloyd asked quick question if we have a 30 second video, for example. It has 30 video frames. Can FLORA break them into 30 frame images?

Of course it can. I don't believe I have a video. Wait. Oh, I do have a video on the canvas because I have my asset library. Let me find a video in my asset library if I can find one.

OK, so let's say you do have a video. If you have a video on the canvas, you can extract frames here and so you can extract all the frames. What you could also do is. Let me see.

OK, so you could extract the frames individually one by one, but this is where something like a custom action. What do you wonders? So you just go to action in this case. I'm just going to generate one. We might actually have an action for it already, but I could just vibe code something. I'm just going to copy. I'm going to say.

Split all frames from this video as image still. I'm just going to ask. I'm going to create a custom action. For those that don't know, actions are deterministic functions on the canvas, so it's something like if we don't have this feature, like let's say you want to split video, like split all the frames of this video into PNG sequences. We don't have that feature yet. You could because that is a deterministic thing that's not using. That's not a generative model.

A generative model doesn't need to split out all those things. You can use a an action which is deterministic thing. To get that effect, so all it's going to do is now just split out all of those video frames and the action will vibe code the feature right there as well. And then also if you're using open and if you're using floor MCP and are co working with chat, you could even ask it to create that custom action. For you. Let me refresh my canvas. And.

Again, if everyone might have joined late, this entire project was created using floor MCP. That's the only way I was able to do this. You guys, I could not do this crazy, this craziness without an agent supporting the vision. This is not possible. FLORA has changed drastically because of our MCP. It is very, very powerful. And it's also very powerful with chat 56.

So you can in the cool part is unlike other MCPs where you connect the MCP and just get they just spit out an image back to you and you don't know how it happened or where it did things with floor MCP. You'll have the canvas. You have the canvas to look back on and see the proof of work and the chain of thought so you can see what this agent has done. It's going to lag very hard. You can ask it to organize the workflows for you can. I had it label everything. I had it test out every single different strength setting.

I had it test out different product images. I had it run through every single various different image model that uses for LoRAs. And it just kept doing it. I think it worked for three hours. I was doing this last night and then earlier this morning. I'm just having a cook up this. I don't know if I can see the time.

But I don't know if I'll be able to see it. But yeah, this is my regular chat if you can see, I just close out that the cool part. That's also why I'm using GPT six now for MCP. I was using Claude before I was very Claude build, but GPT six came out and It's really good. It is really good. And the cool part, especially with the use codecs, use cloud code or codecs, because both of those can actually control your computer while the standard cloud or standard chat GPT can't really control your computer as much. It can't like dig into things as hard.

And so I have, I'm using codecs right now and with codecs, I can, let's say I can close this. You just have a typical chat window. You'll see all my agents outputs and things. But then the cool part is I can then view that FLORA project that we're doing and the MCP right there.

And then I can go full screen even as we were watching earlier. And then you can see I have chat GPT still here. So I can have it, the agent control the canvas. It's like Fauna on a crazier level,. It's really awesome. I've built workflows where it just was look at this latest email. What the client sent me, download those assets, throw them on the canvas and build that workflow.

And it does it and does it very well. It's quite scary. But I think it's because we trained it on a lot of us for deployed creatives point of view and tastes. So we really helped make this a feature incredible for you all.

So excited for you all to dive into using MCP more and MCP can do everything on the canvas. It can create images for you. It can create images for you. It can create edit videos. It can use custom actions. It can create slide decks for you. It can store as a group things together and then save it to your library.

It can leave labels on the canvas. It can organize the canvas. It can color code things. Yeah.

But yeah, it does run through your. I surprisingly know this does not run through my Astra. I still have 65 left. Surprisingly, no, but it does when I'm trying to mix and match FLORA and other various creative tools altogether, because then it's just taking more compute and using different things.

So it's taking a lot longer. But if I'm just using for MCP regularly, it's not too bad. And it's also we've made it very easy for you to utilize the MCP. So if you open up any campus project, you'll see open an agent right here, and then you can choose any of your preferred agents from Claude to judge to cursor to croc to also a more general MCP. If you're not using those, if you're using some co-pilot, you can paste the link and start using the same thing. It should work.

But for the most part right now, I'm currently GPT and codex and GPT six pulled. But I'm sure in a couple weeks, I'll be using Claude as my driver for FLORA MCP, because the new fable, whatever model comes out, it always changes. That's the thing you have to note about the space. It is constantly changing. And the new model comes out and you have to kind of shift and work very quickly to adapt.

But right now it is codex with FLORA MCP, which I really like. Last thing before we get back into to LoRAs, we did drop our official plugin for FLORA. So if you go to plugins, you can if you go to plugins inside of your codex or chat GPT, you'll be able to find our FLORA plugin. So it makes it even easier for you to set up FLORA inside of your chat GPT. Yeah, another example of this particular kind of surface kind of blue sculptural vibe, and then creating more images in that style very quickly. I do want to show actually, oh my god, Lloyd, if you're still here, the custom action did the thing. Oh, Lord.

So yeah, you Lord, you can split, you can use a custom action to split every single frame from your video. Really quickly. I'm gonna delete this because I don't it's oh my god. So much stuff. Delete. I hate when it goes on top of my fields because then it makes it really difficult to go through all them.

Okay, I'll take care of that later when I publish this workflow No worries like Someone said try to use the MCP to have Claude read labels So when I made a change in flora Claude could pick up my changes and update my every motion videos, but I had difficulties with that info Going back into Claude Are using Claude code or using Claude because if you're using Claude code, you should be fine It should be able to relatively simply do that while Claude regular not Claude code Probably suffer a little bit more because it doesn't have direct access to your computer I don't know what it does but like have agents having direct access to your computer make them way more powerful and they can do more things that like I'm still figuring Out and learning and so definitely recommend that Do you mean labels as in like these types of labels Because it should Yeah, you can try again that we've we've made some very large updates with MCP and you should also now see When you connect We open have a project with an agent inside of it You will see your agent as a cursor and you will see them as your name from for me It's a Sasha's agent and you'll see it on the canvas and you'll see where it's touching and doing things and yeah, we've very much improved also the organization the agent is much better at Organizing the canvas that was very much a problem for a while, but now it has been fixed and it's it As you can see it did a great job organizing this project for me today Oh wait the thing I was gonna show so One of my tests Was actually I was very curious because I have this past Adidas Collection imagination thing that I did a while ago inside of flora and I have that project In our community page if you want to check it out is also very crazy And this was also before MCP when I had to when I manually did it all it was a very crazy project I trained Alora on my set of images because I was oh is curious if I could use Alora to create more kind of fashion in this world And I did but what I found for me personally, I only used it on a recraft I should maybe try it on flux and Krea, I know I did try on Krea. I didn't like it too much, but let's recraft Let's go recraft Okay, yeah, there it is synthetic Adidas I have that style let me spawn the other one image Krea LoRA Okay, there we have it. Let me make sure I apply my LoRA And go there let me choose Three by four so I have both of I have two loras of I call it synthetic Adidas Both on recraft and Krea to trained on these sets of images and I'm gonna show you what I actually found that I was Not too I was okay. Well I'm not gonna try to use it for this purpose more.

So Let me use a text node I'm gonna say I'm gonna blow your minds right now Help me create a prompt to help me create a list of prompts Just imagining different front facing shots of diverse models I'm set on a white background with wearing blue black and white abstract streetwear That's inspired by the likes of things like Balenciaga street culture, New York City But there's no text that should be featured on these graphics Output only the list of prompts no preamble separated all by new line Cool just voice detection just using voice detection that create a prompt. This has been my big level up recently that I've discovered. Now I have-- I'm going to create-- I'm going to run that through GPT 6 inside of FLORA using the text node. I'm going to have it create a batch of prompts for me to quickly test out this style on various different prompts so I don't have to sit and think through each different prompt right now. I'll have GPT 6 generate this. I will go to batch.

OK, great. Now I'm just going to separate by a new line. I will plug in my batch into these. OK, make sure my-- OK, my layers are applied.

OK. I'm actually going to duplicate this. I'm going to duplicate this four times because I want to test it out against actually-- I'll duplicate this one twice. Let me test out because I want to test out.

So on the recraft side, I'm going to-- I have my layer applied. I'm going to set the style-- style match to precise. I'm going to set the style match on the right side to flexible. And let's test out.

Let's take a look at how they interpret that. Again, at the top is the recraft models. Let's try the same batch of prompts against the Create 2, FLORA. I'll keep this one.

Let's keep-- actually, let's keep this one. 2.5. I think 2.5 is the good one, right? That's flux. Yeah.

OK, it seems that the Create-- specifically the Create 2, LoRAs are much better at when you set the strings to 1 or 2 versus when you set it to 0.05. It's not as strong. OK, let's get back to where we were. OK, I'll set the Create 2 one up here. I'll keep-- let's do 2.

Let's set it-- no, set it to 2.5. For this one, then we'll do 2, then we'll do 1 and 1/2. And then we'll do 1-- OK, 2. Keep it at 1 and 1/2. And we'll keep this-- we'll keep that one.

OK, so I'm going to run all of these across all of them. I'll have it take a stab at that. OK. So-- oh, man. Well, I guess-- well, let's not do it at 2 and 1/2 or 2 for the Create ones, because they came out like trash.

So let's tone it down, actually. Let's do-- let's do 75 on that one. And I'll do 50 on this one. OK, let's take our-- take a run on those. One second.

OK, I'll let those generate. I see a question. Oya asked, Sasha, what if we have a hero shot and need to apply the art direction from the aura so it is that possible? You showed an example with the Shiny Cat, but you generated that from text.

So the Flux 2 Lora model is the one where you can-- you can see right here. The Flux 2 Lora model is where you can actually have a hero shot and then plug it in and use it as-- to apply that specific look. So let's say we have a model, but you want to do that sort of glittery cat-like photo shoot of that model. That beautiful model picture, you're just trying to apply that style. You can use the Flux 2 Lora if you've already trained it on Flux 2.

But again, Flux 2 Lora's do take a while. So just bear with that. You will get the-- when you're done training the Lora, it's going to take like a couple hours until you get it back. That one just takes the longest, because it is the old-- the Flux 2 Lora is the oldest Lora model available. The current newest ones are the Kreia 2 and the-- recraft ones.

So those are much faster. But yeah, so here we have LoRA trained on these pieces. And I'm just curious to see if I can create a fashion collection, kind of continuing that vein of creative direction that I've established. And let's take a look.

So actually, okay, last time I tried this, I didn't, I feel like I hated it. But I think, okay, once I set the strength to one and a half, and that's when I feel okay, it finally was falling along. But the problem I kind of see is that sometimes it feels a little too close to this, like where it's I feel like I can't, like it's not expanding out enough. But I think it's fine in certain cases, I think I don't think it's actually that bad, surprisingly. That's interesting.

But you can see it's not always, oh, these are cool, actually. I don't feel like recraft did it that well, in this case. But I feel like the create two ones overall did. Okay, so yeah, if you have a collection, they've made or like visuals that you've done in the past, you can, create a crayon, probably, I mean, try all of them, because maybe like a how a different model portrays yours. For me personally, I like the crayon to visuals the most, and then mess around with that strength setting. It's kind of see what happens.

But now you're able to just kind of continue that same sort of, creative direction you establish here, let's say you spent a bunch of time creating all these really cool AI visuals of your imagined, brand, or your imagined fashion collection, or imagined product, or your imagined art, you can train it, and then keep continuing visuals in that same kind of universe. I think especially with LoRAs, you definitely have to just try out yourself. It's very much a personal kind of expression and feel things, but you train yourself and you train a model on your visuals, and then kind of see how each one of them interpret it and take it. And again, you can use the flux to model with LoRA, where the flux to model can take another image as input to generate while we both recraft and create to cannot take in images as input to do something with. I recommend it for like more art stuff.

So a lot of like AI artists that I know do a lot of like this LoRA type of training, because then they like create a vibe. They have a certain aesthetic, certain feel to their visuals, they can create a LoRA. So that way, they can continue that. Because the cool part is you can, you don't need a crazy prompt anymore. Like you as we saw earlier today, I just said a cat sitting on a desk, and I had that LoRA, that glistening cat applied, and I was able to get a visual of that.

So that's also the beauty of LoRAs, because then it cuts out the need to have crazy prompts. If you really have that established, creative direction and that sense of mood. So then you can keep building in that world continuously, and just rely on simple prompts. Let me just delete all these, we can clean this project up. And I will share it out. We'll publish this on the community page. Let me choose.

Community page. So you all have access to this as well. Actually, let me find a cool visual with like a flower in it or something. I have so many.

Let's roll this. I will publish this, the community page. I will share the link in the chat again right now. Feel free to duplicate this project and use it. Keep learning from it and using it. I will share the link in the chat. I will share the link in the chat.

I will share the link in the chat. I will share the link in the chat. I will share the link in the chat. I will share the link in the chat. I will share the link in the chat. I will share the link in the chat. Something to note, there's been a UI update to FLORA there.

As you can see, the canvas looks a little different now. You can change where your toolbar position shows up. I have mine set to the top now because, I don't know, I just like it at the top. But you can set it wherever you prefer. It can be in the left, the top, or the bottom. You can place it there. And now it's in this one seamless.

Very clean little block where you can access all of your techniques or your move and pan to the field icon. You can even search the canvas now. Stuff like that. You can also create folders of your styles. You can create a folder of these ones and put it all in there.

Let's try to create some more. This one. I'm going to just duplicate that. I have this style. It's called loose. What's it called? Loose ink.

Loose cyan ink. Let's, I don't know, create some. I'm going to say flower icons. Let's go there.

Let's go there. Let's see what happens there. Let's also try it on styles pro. Let's say Brooklyn Bridge.

Let's just do that. So I just have my cyan ink style applied to my recraft model here. And I'm just saying Brooklyn Bridge. In this case, I just said flower icons. And, now we're with a very simple prompt and just a LoRA style attached, able to kind of keep continuing that sort of specific look and feel throughout any kind of image I try to create.

So it's really great for brand building because you can, again, train a style slash LoRA that then just prompt whatever you're thinking of and then create a visual set and that same sort of world. But, yeah, I will pause for any questions. I think I feel like I covered pretty much everything. I set the time for two hours of assuming today, but I don't know if we need that full time for today. I think there's all different various use cases or different examples, different, all the different settings. As you can see, there's a lot of different settings and a lot of ways you can utilize LoRAs.

But this project should have a ton of it for you. And again, you can use your agent to help build out anything. You can like how I did. I was here's a bunch of my stuff. Can you go build a floor workflow utilizing all of it and setting it to all the different image models that can take in wars and do something with it and output everything in different with all the various settings so I can take a look at how all the various settings affect it. And it did so if you ever stuck on workflows, just open an agent and tell your agent where you're stuck on and have it build out that example for you. Yeah, another LoRA where we have this like wireframe kind of simulation aesthetic and we can see how I think flux didn't do and do that bad.

Did pretty well. We craft did actually do really well in this case, certain styles and LoRAs I feel like certain models do much better at while others don't. So I always kind of experiment with that just there's something about their training data that likes those visuals more than the other. It's very random.

But yeah, we crafted a great job at this sort of simulation. So I could say something let me, let's use, let's use precise. I will say, I will change the aspect ratio to 3 by 4. And I will just say a lower and a base.

And then now we're going to get a flower and a vase. That's kind of set in this sort of wireframe creative world. Cool. Again, training LoRa, simple prompt, able to continue out that same visual world that was established from the LoRa. Very powerful. It can help expand kind of what you create. I think it's really, especially LoRa's and slash styles are really powerful, especially for traditional artists or traditional creatives that might have a ton of assets, but they don't feel like any of the models are producing it as accurately as possible.

What you can do is then train a LoRa on your style and then create more visuals in that world and really augment your creative process. What's up, DJ? How's it going? I don't know how late you joined in or early. I think we're pretty much wrapping it up at this point.

But yeah, used MCP to create this crazy project that has every single LoRa setting across many different styles that I've trained it on. MCP for the win. Any remaining final questions before we wrap it up for today? Again, I will share the project. It will share the community link to the project again. Feel free to explore on yourself. Feel free to train a LoRa from these images.

Again, it's going to be these. If it's a cohesive set of images, you can train a LoRa on these so you can just download them yourself and then go here to the file, to your asset library, go to styles. Add a style, train your LoRa and then train your LoRa right there. I'll go through it a little slower, actually. Styles, add style, train your LoRa. You can choose which model. Honestly, I recommend trying all of them.

But you can also see my examples that showcase from Kia2 to Flux to ReCraft. They should all be here so you can kind of see a visual sense. But certain styles, just certain models like that style better and you can kind of see it here. In this case, I feel like Kia2 didn't do too well, actually, with this look right here, this grainy light film. While I feel like Flux did much better. Actually, no, I think ReCraft did the best here in this case.

So sometimes I recommend training the LoRa on all of them and just kind of seeing, comparing, contrasting how you what you think is the best for your use case. Yeah. And again, you can use floor MCP to train LoRas for you so you don't have to, think through all the settings. So there's a lot of settings that can feel overwhelming at times, but have the agent help you out. Amazing. Well, good luck on your brand, new brand this weekend. I hope you have fun creating it.

And today's lesson helped give you more skills on how to think through and what you can create. Yeah. All right. Well, with that, have a great weekend, everyone. Take care. And we'll see you next time. I am Sasha Zabegalin, a forward deployed creative here at FLORA.

And I'm signing off. Talk to you later. Bye. You