How FLORA designed its New York out-of-home campaign on the canvas

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

Hi, everyone. Welcome to this week's webinar. This week's webinar is with our very own Nick Miller, head of brand at FLORA, the topic being Canvas to Billboard. We host these webinars every single week and have many more plans.

So please check out our Luma page where our education calendar lives. And on that note, I think we will get started. I will pass over to you, Nick. Hi, everyone. And thanks, Olivia. Appreciate that.

So yeah, today we're going to be talking about how I joined FLORA and needed to do an out of home campaign for FLORA itself and how I kind of through that process, how I had my aha eureka moment, not just with FLORA, but also AI as a whole, which, hopefully you guys find compelling. So yeah, this is me. I previously was working at Apple for about three and a half years. I came up doing brand work, working at places like Gretel, Buck, etc. On New York.

But mostly a generalist, I would say at heart. But brand is definitely my thing. So yeah, when I joined FLORA, I hadn't really I had dabbled. I had definitely dabbled with some AI tools.

But it was definitely it was definitely a bit of a journey. I hadn't really, I hadn't really done it professionally, especially at a company like Apple previously where, a lot of the design tasks are still can still be very manual. And that doesn't change. But the speed and augmentation that I learned through working here at FLORA, but also through this out of home campaign was really eye opening. And I definitely like I can never go back now coming from this kind of traditional design world to now this more AI forward world. I'm sure that resonates with some people out there.

But yeah, I hadn't really done prompting or anything. And so I was kind of thrown in to this whole situation, which is honestly a great opportunity. So today, we're going to talk about, yeah, how I kind of got pilled through working here at FLORA and also through this project in particular, which was a very real project, you're talking about putting up billboards across New York, full out of home campaign, all made partially in FLORA. And we'll go over the whole kind of pipeline today is really going to be more of like a raw walkthrough of not just FLORA, because, we're not going to like FLORA is was a big part of the process, but we're going to walk through everything we're going to walk through. We'll take a look at some of the Figma stuff. We'll take a look at how I was using Claude to help ingest specs from the vendors and then set that up, including like timelines and everything like we're going to be zooming around the campus.

So get ready. Buckle in. This is not just going to be a slide show. We're going to be like we're going to be zooming around.

So really, yeah, this is just about how I learned what AI, how I could augment my workflows, especially coming from, Figma, Adobe Suite, After Effects, and those tools are like they're they're not cutting edge anymore. I would say they're like kind of bordering on legacy now. So it was really a lot to get up to speed on. And again, like I'm sure that resonates with some people out there where, you're not really sure where to start with AI. And I totally get that.

So here's what we're going to be looking at. These are the final out of home pieces that were created around this campaign of it's never been a better time to be a creative, which I especially being here at FLORA fully agree. It's a phenomenal time to be a creative. But so this is one of the first placements we did, which is actually a hand painted billboard on Spring Street and Soho. I believe it's getting it's getting painted right now.

So if you're in New York and are walking down Spring Street, you might see this, which is pretty great. But yeah, FLORA was used heavily in this process. We'll talk about that. Here's another one.

This is on seventh app. Again, it's never been a better time to be a creative. And we're going to talk about, iterating to get the right imagery and, how you can go about that, how you can use FLORA to also do some of the less glamorous parts of production using techniques to resize imagery, etc. And then these are some of the subway placements.

So yeah, we'll talk about how we did the sketch to render effect, which is pretty basic. And it's definitely like just some prompting. But also there's some there's some model knowledge there that I think is good to know. So we'll take a look at these as well.

But yeah, all of the all the imagery here was created in FLORA. So taking a step back for a second, for me personally, where AI tools were kind of not bridging the gap for me, I think that coming to FLORA, I had especially coming from creative tools, I had an aversion to prompting in a way that's I'm just not very good at it. And I don't I don't want to be good at it. I don't necessarily like it is important.

But I don't necessarily want to like become a prompt engineer. I'm sure some people feel that way too. It's like it's not it's not what I'm here to do. I'm here to make great imagery that feeds into great work.

So not being able to prompt was a big issue. And this was this is like my first month of the job. It felt like getting something good was really difficult. I felt like I was banging my head against the wall to, again, like it kind of feeds into the prompting thing, maybe it feeds into some model knowledge.

But it's not, it's not crucial. The model knowledge isn't crucial. But getting something really good, that kind of hit a certain bar for me, felt impossible, truly. And then, with FLORA, like I was very interested in FLORA as a product as well, because it's node based, I love node based tools. I, come from obviously a design background.

But like I said, I kind of like a generalist and a lot of the 3D tools, like I use Cinema 4D and all the like shader networks in like Redshift and Octane are all node based. So I was kind of this is great. But for me, when I started using it, if it still felt a little manual, like it felt like I was this is a really powerful tool. And I'm still connecting all these nodes, which is inherently important.

But also I just felt like I month one on the job was me feeling like I feel like there's a better way. I feel like there has to be a better way. And this is when I'm getting requests left, right and center for on brand FLORA imagery or a myriad of different projects going on simultaneously. And I'm just sitting here thinking, there has to be a better way to work in FLORA and to kind of get what I need at scale, get good results at scale. That's really what I was after. And this is all going on.

And then my manager comes to me and she's shout out Megha, she's the best. She's so we have this project. It's billboards in New York. They're already bought. The placements are bought.

So we're we're doing this. We're doing an out of home campaign. It's like nine placements. And this is we need to just we need to do it. We need to figure it out. And again, this is I'm still going through this whole like month one, like I'm I just feel like I can't get anything good out of FLORA. I feel like I can't get anything good out of any tool.

I feel like I'm like struggling to integrate AI into my workflow. But at the same time, I need a ton of imagery. Like imagery is such a crucial part of any brand. And so not only does it have to have a certain look and feel to it, but I need a lot of it.

So we have this album project. So this is really where I have my moment of FLORA. And this is kind of what this talk will be centered around in terms of FLORA usage. So I we were doing this project, this other project. And one of the FDCs at FLORA came to me or I asked him to help me generate some stuff because I was really struggling. I was we need like a ton of imagery for this project that's completely separate from the out of home. And he sent me a canvas that probably had like twelve hundred images on it.

And it was all perfectly gridded and it was all perfectly laid out. And it looked so good. And I was how is that? OK, I need the sauce. What are you doing? And he showed me the MCP.

So how the MCP works, what it is. If you go into FLORA right now in the top right corner, you'll see an open an agent. This is kind of the secret sauce for me. When you open, when you work with Claude or with ChatGPT to generate things in FLORA, it's this like one level remove where the best people, going back to not being very good at prompting and not really wanting to learn how to prompt. Not having an interest, the best people, the best things to prompt, the best things that prompting our other bots, their other A.I.

This is literally just the case. And I learned this very quickly because I started to play around with this. I connected FLORA to Claude and I had it start to an MCP is great. Like the user experience is really nice and it'll just ask you for a workspace ID and you'll kind of work with it, set it up. You'll fire some test images and then you're good to send as many as you want or need and kind of hone it in. And so I started doing this.

And then on top of that, I also he kind of led me to Krea too. So this is like this kind of model knowledge that's a little tricky. Like to, it's kind of just floating in the ether. Like it's unclear exactly. I just happened to pick it up from him, but Krea has this great model called Krea 2 and it has some resolution limits, but it is it's got that sauce. Like it's got that secret sauce. It's just so like it's it just has exactly what I needed for this project and just like for FLORA in general.

So this was my aha MCP moment connecting FLORA to Claude and having Claude prompt different images for me. So when you're looking at this, when you're looking at this right here, the screenshot, all of these I put, maybe it's not on brand, but I'll and I'll walk through some of the, some of the actual use cases to get into more detail here. But when you say generate 30 images or five images or whatever, Claude is the one who's writing the prompts here. And yes, you can do this. I understand you can do this through the batch node also. Yeah, through you can do this through nodes in FLORA. I just prefer the MCP route.

I feel like the results are a lot better. So this is kind of why I'm I'm preaching MCP here. It is possible to do this. Like we have Fable and we have those models in FLORA, but having it in Claude, I thought it was a really great experience. And so each of these, each of these images are a different prompt.

So I'm not suddenly like writing a manual prompt, nor am I, wiring up nodes necessarily. I'm going through and batch generating. I'm generating a few, maybe five, and then kind of curating from there and honing in the look. And I felt like it was a really natural extension of my work, especially in terms of being like being an art director and just art direction in general. It's art direction is an underrated skill, I would say.

But I would say if there's one skill to index on in this day and age, it would 100% be art direction. Because how I was kind of thinking about these as I was building them was how would I explain this to how would I give notes to, someone like a person. And it actually it really owns that skill and it's a really important skill to have. So how the campaign came together.

This was a pretty typical design process in terms of ideating, collecting references, pitching directions, building the look, which is what we'll talk about more in a second, and then design assembly and print prep. So let's talk about honing in the look. And, what that looks and what that looked like for this project. So the first thing that I did was we needed to we needed to define for Claude and for FLORA what a flower looks like. Because I think that references are really important, but I want to be very specific when I say that. References are important. Never use just one reference.

Never like just yoink something. Don't do that. I put together 150 images of flowers, and I just jumped it into Claude as context, the whole thing. And so you can see my prompt here, like this is when I was first setting up the MCP. And I said, first, I need you to confirm that you can find and read all the images in this older FLORA imagery poll in my documents folder.

So you can I need you to look at and analyze all the imagery in this file for every generation in this session. These will be your references. I will be adding to it. So you may need to cross check for updates per generation. Please confirm once you have loaded the images in this folder as reference for this session's generation work. One note here shouldn't have said analyze. It will actually like give you an artifact.

If you ask it to analyze, just you can either pop that out by saying don't give me an artifact or whatever. But yeah, it's I'm telling you to ingest the context, like really look like look at it. Look at the flowers. So here's our first generations for or just in general for the MCP.

This isn't even this is like this was kind of I was I had the out of home stuff in mind, but this wasn't necessarily like we weren't quite there yet. So in this instance, again, this is kind of plain language in a way where which is again really nice where you can say based on these images create a few flower images on a dark and booty environment, the flower should be a bright color and the environment should be more graphic and studio like than anything you spray it to and here's the workspace ID. So within the MCP, you're going to define your model, aspect ratio, resolution, and then workspace. Pretty self-explanatory makes sense. And this right here was probably I think this was my aha moment because I was okay, like I was getting results that I felt like I could see where this was going. Like I could feel I felt like I could work with this. We can actually get some stuff.

And for the first time since joining FLORA, I was okay, I think I can do this. I think that I think I can figure out this whole AI this whole generation thing. So from there, it's about art direction. It's about iterating and art directing.

So, here's an example of a prompt where I take some reference images from the same folder. I take some of those images. And this is kind of an example of how specific you can get with the MCP and how well FLORA responds. So I said great positive reinforcement. Great. We're using we're going to use these flowers as a base and try out some vibrant colored backgrounds. See the references attached here.

Feel free to use the color of the flowers as a color on color background or choose a complementary color similar to the references attached. I'd like to see nine images as a test overall. Three with rich and vibrant color on color. The color of the flowers is the same as the background. Three with rich, vibrant, complementary colors. The color of the background complements the color of the flower.

And then three with more earth tone background colors to complement the color of each flower. No need to stick to just one flower. These can be an arrangement. The background should feel graphic, but everything should feel lifelike and photographic.

And then the facts of what I'm looking for in terms of resolution, etc. So this is an example of a example of, to me where the MCP starts to feel really natural, where it starts to feel like I am not, in there, like connecting things and wiring things, but I'm actually art directing like a junior designer or something. And this is where I'm okay, I'm much more in my element here. We can do this.

So here's another example of some of the imagery that we wanted to use. So this was, I had already built up a bit of context and had pulled some imagery like this. So this is kind of like a few iterations down from the main image. And this is why, there's only one image that I'm referencing here, because I had already referenced a few above, but I pulled this one in particular. And I was look at this image. I like the idea of somebody who could be a male woman laying in flowers, daydreaming with this kind of old school grainy aesthetic, but also moody and nostalgic, almost melancholy, and then generate three images in this or generate three test images.

So this is where we're really starting to hone in the look and feel. So when I presented, when I presented the out-of-home work and kind of, we were talking about it internally and kind of, selling it through as you do in a design process. This is one of the kind of, this is one of the looks that we were really gravitating towards. We really like this idea of like whimsical nostalgia and making it feel kind of like familiar, but also like old but new,. And it kind of ties into this idea of it's never been a better time to be a creative, never let a good idea go unexplored. One of the lines on one of the billboards I didn't show is it's never been a better time to be a dreamer. We actually used this image for that billboard.

So this is from that billboard came out of this batch of images that I worked on with Claude and in FLORA. So I was really happy with some of the results here. I thought they were really great. Cause again, all the flowers in here, it's a bit of a black box, but you can assume that contextually Claude is and Fable specifically is going to pull from that folder of 150 images as well.

So you're kind of building up this context, right? Of here's what a flower looks like. Here's what, I would say is like a nice looking flower versus just like a flower, here's like an elevated version of a flower. And, it's again, bit of a black box, but for things like this, like images like this down here, these kinds of fields, I would assume that it's also pulling from that context.

So that's kind of the beauty of the MCP as well, is you're building up your own context as you work. So, and that kind of informs future generations and helps you build a look. So what the MCP got me, or me personally coming in very cold in a general sense, it got me speed. I was able to generate imagery at scale very quickly. It got me quality because I was able to art direct and part of that had to do with model usage, yes. And like what models were used, but it got me quality quickly.

And then what's interesting and just like an overall note about kind of creativity and in the world of AI, which, ties back to what I was saying earlier about art direction. My job became very much about curating from nice looking stuff. Like you have to have some form of judgment and taste and an eye to curate down from 10 images, 20 images, 30 images down to like three. Like I would maybe choose three. And again, not phenomenal for, this is usage intensive, but this is an intensive project,.

So I needed to make sure that I was getting what I needed. But I do think that moving forward, I think that the judgment layer for us designers and, artists and creatives in general is only going to become more important. This idea of, working and generating imagery and videos and, creating and designing as it becomes more and more AI assisted, it's going to be more about, okay, well, why does this work? And why does this other thing not work? And that's a very important distinction. And that's something that is the key to, I think in this day and age is taste judgment.

Okay. You like an image. Why do you like an image? I remember when I was first starting, like I would do that exercise with myself, just in general, like I would try and dissect why I like things and why I didn't. And I found it very helpful. And I think it's only going to be more helpful.

So this is kind of what changed for me. This was my aha moment. And then let's talk about, and let's go through the full pipeline. So I'm going to zoom out of this Figma file and we're going to go on a bit of a journey.

So in terms of the full out of home pipeline I'm going to just show you a little bit more about how we, like what the out of home, what the out of home pipeline was, but like from the perspective of like the whole production, the whole thing. So let's, let's go in and take a look at, let's take a look at, so I'm just going to give you like some behind the scenes for a second. So when I was talking about selling it through, so this is like our out of home campaign. This is we're doing creative directions.

So, we're talking about, this is was one theme that we could work with some references, this idea of it's never been a better time to be a creative, some kind of sub, some sub concepts within the theme. And this is mostly for copy. Again, we won't dig too deep into this, but just to show you okay, here's how we were working and then early content thing. So this is again, just maybe even like bashing in reference, but then like you can see I'm in the middle of this process and I'm starting to use some of the imagery that I was just showing. And you can see that I'm using this in the pitching world to kind of help bring ideas to life. Again, that's what this is all about. That's just what, that's what the out of home campaign is predicated on.

That's what, that's how FLORA empowers you is you can just bring it to life. So you can see I'm starting to work here and just like some really rough concepts, right? And these Floras were all again, this is like baby steps into the MCP. So this is kind of how we were pitching it.

So that's one, that's one kind of area. And then let's show. So this is the first out of home. This is the first out of home canvas that I was working on.

So this is when I was showing you the prompts, this is where I was generating into. You can see those initial prompts, initial images here again, like these were heavily used. This was used. This was used again, curation. If you zoom out, we did a ton of exploring here, a ton of exploring, but like these images were just, this is where I really again, just continual aha moments of this looks great. Like this is such cool and interesting stuff. Yeah.

Some of it's a little bit more AI, some of it's a little like works better than others, but you can really control and art direct the look and feel. And also like now you can get even more controlled with this stuff. We just launched 3D node, which is now in FLORA. And you can use, again, you can use the MCP if you want, or even on the canvas to like build, you can build components, 3D components, you can build scenes and lay them out and then render them all in FLORA, which is really great.

So this is I mean, we generated a lot of stuff, a lot of stuff, but it was necessary. And this was, I think like I was still learning how to use the MCP effectively. And sometimes you're going to get better results than others. Of course.

This was like just alone for I was trying to figure out a better, way to do just these two, just these two process images. And so that was like a struggle. And it's always that's getting those details, right? Is always going to be that's always going to be really difficult. Like I find that those types of moments in the creative process, those like details are what make things, that's what makes things great.

So yeah, sometimes it's going to take a long time. We were looking at the whole idea here was that these two should feel like they add up to this dress. And so we were doing, I was doing explorations around brellas and, just all this different stuff kind of, again, like within the MCP able to, I'm able to retain the look and feel that I had used previously for other generations, because of that context that I had built up. And there's some really beautiful stuff in here we ultimately didn't use. These are my personal favorite. I think I asked.

So I, this was from one of the flower generations. And I thought it was great. And I asked it to generate just riff on this because I wanted this flower, but like closed and from different angles. And it was and it just did these, I thought, just really gorgeous, like flowers that were just so awesome. I just love these so much. They feel like somewhat painterly. I don't know.

I just, I love this stuff. So this was, again, like just aha moments left and right through this whole process. And then in terms of getting the, let's quickly touch on. So this is one, that was one. Canvas, this is another canvas again, pulling in some of the selects.

This one's like even more kind of raw potentially. But so one thing that I had to do just side note or the subway ads, our hero imagery. So this is what credit to, this is what generating credit to, which I thought was really awesome. It's a great, it's like floral exploration. I loved it.

But it's obviously very washed out and it's also not very high resolution. And these are going to print and they have to be a certain DPI. So using the technique image upscaler, when you're in FLORA, double click, go to techniques and search for it's there. It's really great. And it will take your exact image. It will add a little punchiness, which is really great.

But it will like recreate it verbatim. So using the image of scaler to upscale imagery, and then using the layout resizer. This has got to be my most used technique in FLORA easily. Layout resizer, another technique. Go check it out. Go check it out in FLORA.

But it will just give you like different variations at high resolution for different cuts, depending on what you need. So I ended up using, I believe this one. So that's one example of using techniques in FLORA. And then I also want to touch on briefly let's touch on some of the sketch to render feeling things.

So this was again, like some model knowledge and a little bit of trial and error. But what you can do is you can feed, again, Claude has all the context of the images that has previously generated. So if you go in here, I mean, it's a little clunky, but if you go in here and I would just screenshot, like this is literally the sketch to render. Of the image we were just looking at. If you just screenshot this, it will know what the prompt was and what it will do. And you can just prompt it and say, Hey, I want this exact image as a sketch. It should feel like it was traced.

It needs to be identical, but it should feel hand drawn. And the beauty of another kind of like side effect of the MCP, which I find really helpful, is if you don't specify a model, you can also just either leave it blank or say, choose the model that you think would be best. I believe I left this one blank. I don't even know if that was on purpose. Cause usually I would just like send out and cry to, but it did a GPT Image 2, before 2.5 is out and it did great. And I was Oh, GPT two is really good at tracing and image. Cause they have image capabilities.

I know that, again, I was not. Pilled like that. Like I did, I would just did not have that innate knowledge. So kind of the same process here where I would go in a layout resizer again. I mean, I use this thing like an addict. And I believe I use this one.

So that's how I was using FLORA to kind of generate some of the, some of the imagery, some of the sketches, et cetera. But then let's talk about, let's talk about the full like design pipeline. Thing is happening in FLORA. These are four print files. These need to be to a certain spec that we got from vendors. I also am not going to iterate personally. I want to iterate in Figma from a design standpoint.

So I'm just going to run through briefly. I won't like show any chats or anything. It's not really necessary, but, I will just show you briefly, like an explain and talk over like how I used Claude and Figma, and Photoshop to kind of like do the rest of it in a way. So this is the out of home thing. Apologies. There's a lot here.

So let's talk about this, this guy, I did not make this was, through Claude, what we did. He's like our creative ops producer. We, he went through and took all the specs from the vendor and fed it to Claude and said, I want to write into Figma and give me like an interactive timeline of when everything is due, when everything is flighting. And it was really helpful. There he is.

So this was super helpful, obviously. Cause then we can see like we he went through and color coded everything in Figma to see, so we can make sure that, placements that we potentially were going to reuse. We're not going to play at the same time. So this is really helpful.

Then we took, and then we had Claude do make these like cards. So this is like great for project management. We just had like gave Claude all the information, from the vendors and said, can you like make this in Figma and make it digestible and like make every, like pull every spec, et cetera, from like the lookbook, et cetera. And it did that. And it was really great. And this is how we kind of like color coded everything and just like worked through it.

So this was really helpful. And then from here, what I did. Is I took all of these specs and I told Claude, I said, Hey, I need everything. I need every placement with bleed crop, et cetera, completely accounted for as quarter resolution, quarter scale documents of Figma. And that is what this did.

So helpful because then I could just go in and I know that this is exactly what this placement is going to be. And so I can just start like designing. And then just working in Figma. And that's what I did. I mean, you can see the process just for that one spring street was this, pretty much.

So, and I'm like using the templates that I, that Claude had made over here. So that was a very helpful thing. And then in terms of prepping files for print, so kind of on the other end of it, once I had generated, finalized the design in Figma, doing all my presentations of Figma, et cetera. And, I'm using the MCP and FLORA. Everything needs to be rebuilt in Photoshop pretty much. Sometimes vendors want in design files or illustrator files, which is not a problem. Why?

Because again, leveraging AI here to do the work that I previously was not excited to do. I would tell Claude after we had like finalized design, here it is. I would tell Claude, okay, I wanted you to go in and in Photoshop recreate this composition in Photoshop to spec in the live file. Actually at first I didn't even do that. I just had to create the Photoshop file and then I would do that.

And then I realized wait, why am I doing that? If I could just do that. So I had Claude, for all these, I would have Claude pull in the exact working, like the exact type specs, et cetera. This imagery, all like the fullest resolution. And I would tell it, can you I need you to flag if there's any resolution issues at the DPI that we're working with and at the scale that we're working with. And it would go through, it's a, it takes a while. Like it's a, it's a complicated task for sure.

But it would rebuild to spec all the files for print. And once I had kind of gone in there's always a bit of, manual tweakage. I would tell it, Hey, I need this to be as editable as possible. And it would go in and recreate these things with live type, smart objects, et cetera. Yeah, could it match the like trace dress to this perfectly? No, like I had to do that at the end, but that's fine.

And then from there, I would have Claude once I finalized it initially because I wanted to make sure that I was doing it correctly. I would go in and manually, in Photoshop, do the print pass myself, but. Because, eventually I, once we got the specs down and I knew Claude was like working correctly, et cetera, I would just have it package everything. So I'd have a package zip. It would get do bleed marks, prop marks, like everything press ready. Press quality PDFs ready for print. And I would email it and that was it.

So that's, that's everything. You