How the FLORA MCP puts Claude to work on your canvas

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

Hey guys, welcome to this week's FLORA Lab. Today we have Catherine Chung and James Kelley getting creative with the FLORA MCP. We will get started in just a minute here. It looks like a few more people are coming in. Pretty big group today. Just a couple of housekeeping items while we're ready.

This webinar, as always, is being recorded and will be posted to FLORA's YouTube channel after the session, as well as sent out to all registrants to this session. In that same email, if you did register through Luma, we'll be sending out a code for a free month of FLORA Pro. So if you're new to FLORA, this is your chance to try it out for a month on us. We host these webinars and labs every single week and we have many more coming up.

So please do check out our Luma Academy page where all of our education offerings live. I will put a link to that in the chat shortly. And then lastly, please keep all questions for Kat and James in the Q&A box rather than the chat so that we can make sure that they are seeing it throughout the session. All right. Looks like we are good to go. Over to you guys, Kat and James. Awesome.

Great. Welcome, guys. Thank you for your time today. Just a quick introduction. I'm sure you guys know us at this point, but I'm Catherine and I'm a forward deployed creative here at FLORA, mainly with a graphic design background. And now here at FLORA, kind of helping with educational stuff like webinars, as well as a customer like tool adoption workshops too.

But I'll let James introduce himself as well. My name's James. I've recently joined the FLORA team as a forward deployed creative. And I've been recently spending a lot of my time defining new use cases for the MCP. And I'm really excited to be a part of this presentation today. Yeah. I, we decided to kind of do one of these about MCP and I was James, where are you at?

So here we are. But I like we have a little agenda for today. This is going to be much more of like kind of like a show and tell just because of the nature of this type of feature, it would be a little kind of like complicated to like have this be a little bit more collaborative today. But we're we have a lot of really cool stuff to kind of talk about and show you guys.

But you can come into this with absolutely no background on what MCP even stands for as an acronym. And you can also come in here with like as much MCP knowledge as James does and still kind of garner something valuable from it. So with that being said, I'll kind of go ahead and get started. But again, welcome to our FLORA lab on MCPs. If we haven't met before, here's a little introduction of who we are.

So FLORA is a node based creative tool that brings the top creative AI models across text, image, video, audio and 3D into one intuitive interface to accelerate creative production without compromising on craft. And FLORA as of today has a few entry points. We have the canvas, which is a node based infinite style canvas that allows you to visually program repeatable workflows, freely control how media interact or generate and allows for a more hands on experience. We also have MCP using a third party LLM or language learning model application to use FLORA to execute workflows and generate images with existing context already built out.

And then we have create slash generate, which is a chat interface that allows you to generate images using any model we have on the platform. But of course, today, we will be focusing on the MCP. So today's agenda, we're going to be going through a few things today, it's going to kind of be broken out into about six different sections, including a Q&A at the end. But first thing we're going to go over is what is MCP? And what does it stand for? The next thing is going to be understanding the different ways of using it.

And then what can it do in FLORA? And then James is going to walk through a few use cases that he's that he has ready to kind of present. And then after that, will kind of be me live demonstrating how I would use MCP or kind of just kind of getting to the concept of it and demonstrating how that might look for someone that doesn't use it as often. There's one thing to kind of know is that I do a lot of node based kind of canvas work, while James does a lot of kind of MCP work.

But at the same time, like to his credit, he does a lot of really amazing like workflow canvas stuff. But it'll kind of be interesting for me to be okay, let me kind of demonstrate this live and ask James for some advice and best practices. And then after that live demonstration, we're going to have a kind of like follow up, kind of like closing out with some best practices to keep in mind when you go out into the world and start messing around with this on your own time. So with all that being said, what is MCP and what does it stand for?

Okay, I like for one, am not a very technical person. I don't think I really I think I like learned what the acronym stood for a little bit. And then like I forgot and then yesterday I kind of learned what it exactly it stood for again. But I'll kind of walk you through this as kind of plain terms as I can.

But MCP stands for model context protocol. I know that sounds like a little scary, but starting with the M model is the AI you already chat with. So think of Claude, chat GPT and tools like them. So now if we look at the C, I'm like walking through the alphabet, it means it stands for context.

So context is everything the AI can see and use. So your files, your apps, your flora canvas, the more context the AI has, the less it has to guess. And then the P and MCP stands for protocol. So it's almost like it's the agreed way that they plug into each other.

So think of it like a USB C port, one standard plug. So any AI can connect to any tool that supports it. Put it all together and MCP lets your AI assistant step into other tools and do the work there for you, not just talk about it. So let's look at the creative process before and after MCP.

So before MCP, you would kind of carry everything between tools. So you would download it here, upload it there, copy a prompt, paste the results back and every handoff runs through you're the one switching through tabs. You're the one switching through windows, all that sort of stuff. But after MCP, you ask and the AI does the running.

So one request and the AI goes into your files and your flora canvas to fetch what it needs to do the work there. So what this might look like is me going to the AI, for example, let's pretend this is Claude and I'm hey Claude, can you go into my files and take the files from Burger Heaven, social media assets, 1.1 and then kind of flow them into flora so that I have those assets to kind of generate off of. So Claude will kind of do okay, I can do that. Let me just go ahead and run that for you.

So you don't have to worry about anything else. You can go take care of something else that you want to enjoy doing. So it's kind of just like a extra pair of hands and really kind of expedites a lot of the creative workflow, I should say. - The orchestration. - The orchestration, there you go. So then after that, understanding the different ways of using MCP.

So there are two ways to kind of like use this. There's hands off, which is remote. You ask from anywhere. So you could be walking down the street. You can open chat GPT or Claude even on your phone and describe what you want.

And then the AI works in flora out of sight. It opens a fresh workspace that you never see and it's great in general for quick generations when you're away from your desk. But side by side is the one that we're gonna be focusing on today. So let's say that I'm sitting here and, or me and James are sitting here and what we're looking at right now is the screen. And on the screen, we have a flora canvas. And we're gonna go ahead and take a look at that. We also have an AI sitting inside the desktop next to the floor window as well.

And the AI builds where you can see it. So every node it adds shows up on the canvas as it works. You can see the workflow start to kind of build right in front of your eyes and you can step in at any time. So you can leave a label like build here. You can tweak a node or you can take it over entirely.

But again, today we're going to be just talking about the side by side way of working with the MCP. So it's where flora shines. You can see every step, steer it as it goes and keep the canvas afterwards. So everything we show from here on works this way.

So what can MCP do in flora? So it can make things for you. So anything that you can do yourself in flora, MCP can do it as well and more confidently. Since it knows how to kind of really untap the higher potential of the tool itself.

So you can ask in plain words and it picks the right model for you for images, video, 3D and audio. And one request can turn into dozens of variations or every size you may need. So that's one use case. It can use what you have.

So your brief, your references, your brand, you can hand it files from your computer. It puts them on the canvas and uses them as context so that the results look like yours. It builds workflows for you as well. So you can describe the process. It lays out the nodes and you can get a real workflow on your canvas that you can open, tweak and run again next time. And it gets you unstuck.

So pointed at a workflow that isn't working. It can explain what each node is doing. So if you want one of the best practices that we're going to be talking about later and one that James kind of emphasized to me today was if you want one of the best, if you want a very good run with your MCP, sometimes even building your best attempt at a workflow and just asking the MCP to kind of take that as your creative brief, understand what I'm trying to do and kind of go from there and make something better is also very valuable. And that's a very valid way of kind of using it.

But James, I don't know if you have anything else to add in regards to how MCP works with LoRA. No, I think this is a good transition point to get into the three video examples that I'm going to, I want to just run through them, press play on them and then walk back through them. And then I have one final demo, which would be a live presentation. I believe like that's the next slide, right? Yeah.

So we're going to go straight into like how to get unstuck with one of my favorite use cases of it. So just let it play without talking first. Cool. So yeah.

And then it's going to just conclude with an audit. And now I guess I'll kind of just like cap off what really happened here. So this all started with, oops, no. I got you. I got you. Oh, I got to be right on here. Wait.

There you go. Yeah. So you can scroll through. Yeah.

So it all started here with a workflow. I had built by hand where I combine a text node to generate and a batch node to generate 12 views of a ecom campaign that the photographer forgot to take the shots of. It came out pretty good. I was satisfied with it, but I felt like the proportion, the proportions and scale of the shoe could have been pushed further. And also like the creativity behind how the model was repositioned needed like some working.

So I used our open an agent feature. You could just keep going. To place this annotation on my canvas. Once I labeled the, put the green label where I requested the fix, I selected open an agent from there. Claude analyzes your canvas. It has full chronological order of all your outputs and can read the prompts behind any output and help assess what you were doing and where you left off.

So its first response is a breakdown of where you're at with your work on the canvas before you even define what your next goal is. From there, I requested it to rebuild the workflow. It took an approach of not using the batch node. Regardless, it got me wonderful outputs. And I actually felt creatively inspired by the output to generate it. And from there, I asked it to pick the best frame to use to create an advertisement. You'll see first thing we did was define the frame that had the most potential here.

I used a color label to highlight it. The Claude can also like it's updated every time you make and you take action. So it now knows that this one's labeled in green. So then from there, I go back into the chat. It read my green annotation. And I said, let's recreate this with the frame we've selected.

So from there, what's actually amazing is the first run of this failed. The club was able to define the why behind it, why it failed. And it was because the person was perfectly center framed in the image. So it recreated a new still where the edge of their foot is out of the frame so they could properly exit the frame.

So I did define details that I would probably iterate on for like 45 minutes to an hour and get creatively burned. I might have paused my project, but the MCP just allowed me to power through and get to my idea a lot faster than I could without it. And yeah, so that's that example. And yeah, it could also extract or input anything onto the canvas or locally to your computer.

This one, I'll just talk over while we go. I find this one really fun because I needed to do a bunch of resizes that included aspect ratios that we don't offer in FLORA. And it was able to generate all of them without me even fully understanding how it did it. I just got the output I wanted. All these aspect ratios are for like arenas and stadiums, giant billboards. My background is in product design. I just got what I wanted and packaged it into a PDF.

So that's just like a simple example of like the ease of solving difficult problems that might take a day or two of research and just doing it effortlessly. Yeah, I could sympathize on having to make 20 different aspect ratios from one design. Yeah, pretty brutal. That's a good way to put it.

And then, okay, after this, we'll transition into like a live walkthrough. But here, I started with generating, taking an image I had generated, turning it into a 3D model. From there, I went in and selected open an agent and used, in this case, I'm using Codex with Astra. Works the same way we are going to use Claude today. And I had it send the 3D model to my printer based on a scale I requested. And it actually followed that up by editing this entire video.

So all three of the videos I presented today were edited by the MCP. I don't know how to edit videos. I also didn't see the timeline editor one time through the 30 revisions I did of this video. It's kind of like vibe coding for video editors. It's really like fun unlock that we've discovered this week. And it's been helpful to like all my colleagues and like myself. I can't believe like I've been able to unlock motion graphics at this level.

In a matter of like what, like two weeks? I know, seriously, you're two weeks ago, I couldn't edit the video. Tell me how I spend like a year just trying to get used to key framing and After Effects. And James is just pulling out videos like this. And it's I made this via MCP. And I'm what do you mean? Right?

Okay. All right, awesome. So I'm gonna let James demonstrate or I'm gonna let James share his screen to show a much more in depth kind of example. But let me go ahead and stop sharing.

And then let James kind of get set up here. Ready? Yeah, we're good. All right. Yeah, you're good. Yeah.

So I just wanted to start with I built by hand. And I think that you will find some like wizardry level wow moments, if you take the things you've built by hand in flora, and then open them in a harness like clodder codex. It is surreal. So in this case, I took I have a sketch of like my logo. And I was trying to eventually turn it into a 3d model. And from there, we one thing that's tricky about stable diffusion models is that they're not very good at physics.

So I actually would prefer to take the models that I generate in flora, and use a software like blender to animate them into like seamless loops. So that's just a preference for what I'm about to present. But in this case, I took my sketch, I took a reference that I wanted to use to define like the lightweight and like the grittiness of the aesthetic I was going for my logo. So I took these two images, I put them in a text node. And I said, I want to turn this sketch into a clean black vector logo. The key weaknesses of my sketch are their proportions and the line weight. The reference defines the graphic style, I want my logo to mimic.

So from there, click generate twice. One cool thing to know about stable diffusion models is their outputs are sticky. So if you don't get exactly what you want the first time, but you want to see a couple of variations and you feel like you could get there. The second click you see in the first click, it didn't quite like adjust the line weight, like I felt like the spacing was still off.

So with the second generation, it like just felt a little bit better. So yeah, sometimes the stickiness of a model could like improve what you want to do really quickly. But on a higher level of creative control from this point, I wanted to see three different levels of line weight to confirm like what I want to go with before I turn into a 3D model. So now I use a text node to say, write three prompts each with different levels of line weight, deliver the prompts line by line, nothing else. Just like this final few words of requesting them line by line, nothing else allowed me to get this output where they have bullet marks.

And then I connect those to the batch feature in flora. From there, I separate them by line. I connect this to an image model. And most importantly, I bring back my reference image and I hit generate. That led me to this result where you see three very different line weights. I right clicked on it, extracted all my outputs onto the canvas for review. I'm just going to delete those for now.

And that will got me. That's what got me to this moment. So from here, I pulled out a node from this is exactly what I was going for. I wanted to thicken the line weight. If I wanted to get very technical with it, I could have requested like stroke sizes, like 0.25 of an inch. I could have requested half an inch. You can use all of your design language you've learned as like a leverage point to create things with AI was trained on like the logic of Illustrator and Photoshop.

So that information carries over and it's still very valuable. And yeah, once I defined the thickness, I liked pulled out a 3D node, no prompt, I had generate. We offer a couple of different models. I actually went with high 3D for this final output I got. And from here, the wizardry begins.

So I selected from this top right corner of work in clay. And I'm not going to click it because when you click it, you're going to get the stereotypical Chrome browser pop up. And then it will take you into a fresh Claude code session. I'm now going to break down the full session of all of the other things I did across other tools.

So from here, we will go into my Claude. And yeah, I guess we can just take it from the top. So when you select open an agent, it will launch this pop up in your Claude code tab. It will immediately start working. You'll get this response of a skill we designed that opens the project immediately and doesn't send you to the floor home page.

So you're like just, you now have a browser view of your project and you will communicate with all the actions you do in floor going forward through Claude in their chat box. And from here, my favorite part is actually before I say anything, it tells me what's on the canvas. There is 11 nodes in one workflow that runs left to right, building my JK monogram intro to a logo. It told me my inputs of the orange marker I sketched. It shows like the cleanup I did using the Astra node that wrote the brief and the design instructions. It also defined the model I used to get the output.

So it's very articulate. And then it goes all the way up to the point where you've even generated the high 3D model made with the bold version. It even knew I chose the bold version and saw the other line weights were thinner. So now it asks me, there's two ways to continue. Do you want to make this a real value? Do you want to make this a real value? And I don't want to do either of those.

I just tell it what I'd actually like to do. So my request was download the 3D model, open it in Blender and create a turntable animation. And it did exactly that. So Claude then went into Blender, generated this scene.

This big sheet that you see in the background is a backdrop and it allowed me to make it in the background. It allowed me to get a white background behind the 3D animation that I rigged up. I didn't have to request that. It just knew that was probably the best way to go about it. And now I can see, you can see the motion here. Yeah, it rigged up my 3D model, perfect seamless animation. All of these squares that you see with key lights connected to them define how the output was rendered.

You can see the final output here. Once Claude generated the Blender animation, it then placed the output back on the canvas for me to continue working with. When we get too ahead of ourselves. Yeah, it also provided me the output in chat for me to review. It showed me where the deliverables were on my computer. And my response was, I know where they are on my computer, but I actually would like it to be uploaded to the canvas because I planned on working with it there.

So it uploaded on the canvas, placed it. And what I actually wanted to do at this point was take the 3D animation and apply after effects, like texture to it. Not a stock one that I created custom. So I went on Arena and I found this reference of this data mashing animation. With just the link and the request, can you recreate this animation and apply it to the 360 animation we just created? It went into After Effects and started working. Oops, sorry.

It's kind of distracting, so close that. And yeah, so when it started building the custom data mashing action, it provided it to me in the chat. Pretty good, but the transition of the outro didn't really meet my standard. So we can go ahead in After Effects and see that I was able to fully view everything it did. And after I fixed the transition, you can see it gives me all the shots and stuff. I actually realized, damn, I didn't even want the white background and I never specified it. And so I just said, can we remove the right background?

What makes this even more impressive is to do that. It had to go all the way back in Blender, recreate that animation before going back in After Effects and doing the new version. So like fully walking backwards between programs, then doing all that decision making, defining how it came out. And I'll step back into the canvas here before we get into the So all of that got me here, where I had a 3D animation, data mashing animation of my logo on a transparent background.

So I used FLORA's timeline editor and uploaded this gradient I made a couple of weeks ago, rotated it and placed the logo on top of the background. This effect is amazing because it has transparency, so I could see the image behind it. And I really didn't expect it to come out this clean. I impressed myself with this. And while we're on the topic of presenting stuff in FLORA and using our editors and all of our amazing features, I did want to show you that I went from the 3D model to our Dex feature, where you can present 3D models and manipulate them live. And you could also put it on an auto rotate effect, so it just like toggles and rotates. Super cool for product design.

I could scale it up. And all of this is fully controllable while you're actively presenting a deck. But we're not done with the wizardry. Going back, I still was like-- imagine I did this in like-- I think this took me like 30 minutes. And I'm now I'm really having fun and I'm creatively inspired.

So I went back in here. I took the first frame by clicking Extract Frame. I took the first frame of this video. I used our Draw tool that you can find under Actions. There you go.

And then I used Draw tool to decide-- at this point, I wanted to turn this into a keychain. And I thought, oh, I could make a mark where I want my keychain to be and then erase a little hole in it. And two ways I could go about this. I did it a little cleaner the first time. I annotated it how I wanted.

And then I realized, could I just do this fully in FLORA? And I was almost able to-- I have no control over the scale of the hole or the thickness of this model. So I don't really like going this direction with it. But it was fun to see I could really rapidly ideate on a keychain and put a little piece of wire and string. And yeah, I've made a keychain fully in AI, starting with a sketch. I made on a piece of paper. And I edited the model without ever touching a modeling software.

But I wanted to do a really controlled, strong final version of this. So I took this annotated, drawn on output, went back into Claude. And from there, I said, can you add a keychain loop based on the annotation I made to this image? Once you're done, send it. I added it to my X2D printer and print it at two inches in scale using black filament in my AMS. It's in slot three.

So those are the type of things that you need to give your Claude or codecs context about so it can one shot execute multiple actions at once. In this case, it's first going to edit my model in Blender. And then it's immediately going to download it, and then it's going to load it in Bamboo Labs slicer, which takes a 3D model and prepares it to be manufacturable. So yeah, two operations across two different tools. First thing it did was fix the model. I didn't even see it fix the model. It just jumped straight into Bamboo Labs, scaled it to the two inches I requested, used the black filament, asked me if I could confirm that it should send it to print.

So I had a preview of my file, a printed mesh, and my printer was running. And this keychain only takes nine minutes. So it was done really quick. After all that, I needed a synopsis of everything I did because I had just done so much.

So I used my FLORA motion graphics skill that anybody could create on their own using any motion graphics skill. That's the reference you can find on the Internet. And I said, the print came out great. Can you make a video about our entire process? Include dupes of the screen recordings of the tools you use, including the FLORA workflow I made, our experience in Claude Code building all this, the steps of creative direction I provided, you animating the 3D model in Blender using After Effects, removing the background and rebuilding in After Effects, then adding the whole for the keychain. And finally, use the footage of the time lapse that I uploaded with this prompt to use to create a conclusion of our process, build up to the conclusion with the decision making and chat of me telling you to send the print to my X2D and running it based on the color I requested. And first, it showed me how it made the changes, what changes it made.

The text on my logo was actually not touching. So to even make it printable, it had to add mesh and connect my two shapes. From there, did a little bit of editing. I actually didn't give it any feedback. And this is what I got. Damn. Just give me one second.

I'll pull up the video. I figured something wasn't going to work. So I'm going to stop sharing my screen, pull that up real quick. All good. Because I do actually have a few questions that kind of like was running.

So how often do you like reach through or read through what your agent is writing down during the process? Because like I know when you compare the amount of chat bubbles that the actual agent is kind of like sending over to you and then the chat, like the one sentence kind of responses that you provide the agent with. It's almost kind of comical to see like either clutter codecs kind of going back to I'm doing this and that. And the reason why I'm doing this is because of that. And I'm going to do XYZ in one, two, three. And you're just okay, great. Do this now.

But how often do you kind of find yourself curiously, like wanting to like read through the process step by step to actually pick out like what it was doing? Maybe you would do it differently, et cetera. I guess it depends on how articulate I am in the tool that's controlling with After Effects. If it didn't get the animation right, I would almost immediately consider just using a different tool because I don't know how to get deeper into After Effects.

But when I'm in Blender, I know so much about what's going on in Blender. I'm going to read everything about what it's saying, give it critical feedback, give it counter arguments of different approaches it could take that it's not considering. So like your level of understanding of a software, I guess, defines like how deep and interested you are in what Claude is responding to you with. Got it. No, that makes a lot of sense. Awesome.

Okay. Yeah. So this was the video that Claude made to round out the entire process of what we did. And you like reminder, James, before doing this, did not have After Effects experience. Yeah. And so he just asked Claude, "Hey Claude, I need you to do a how-to video on everything that you just did for me." Because I don't know how to do it. Which is crazy to me.

Right? But like to that effect, like you so kind of had to ask Claude to "Okay, these are the parts about this process that I care about that I want you to highlight in this video." Like you still had to kind of go about like giving it as much direction as you wanted to. Like focus on like the data mosh and then focus on the kind of annotation of the like key chain loop and like the measurements of it and how like the letters are separate and how like you added the etc. So it wasn't just "Okay, now make a how-to video." You like took the time to kind of like just define the parts about it that you wanted. Yeah. Very, probably the most basic level of scripting possible. Just like here's a reminder of the steps that I value that I want you to communicate.

Now use the video editing tool we've made. There's a question here from Gem that says, "Do how Claude is doing this video in FLORA Timeline or in apps on your Mac?" Yeah. So the actual editing and creation of the video is done with Remotion. It is a code-based video editing software. And that is what Claude is calling on to do it's an open source tool. It's not something you have to download it on your computer. It's also not something that costs money.

Nice. Let's see. If you don't mind. We can switch back. Awesome.

Okay. But there are quite a few other kind of valuable questions I think we can spend like a few minutes on. Oh, yeah. The most important part. The key chain. Real key chain. Yeah.

That's a physical key chain. People. There you go. That's always so mind boggling to me, man.

Let's see here. A few other questions. Let's see. Ooh, this is a good one. What's the best way to manage credits with MCPs? The best way is to start a project by saying, "Hey, I value like my credit spending.

Here's what I would like to do. What do you think it's going to cost? And what's the best approach to spend my usage across all the platforms I work across most effectively?" Yeah, definitely. Because I feel like especially with something like MCP, it's easy to kind of just "Oh, well, if it's easy to just get a hundred variations of this one thing, because I don't have to do it myself," then it's kind of easy to just like let it rip and go.

And then all of a sudden, "Damn, my usage is gone." If not, I've gone over quite a bit. So exactly to James's point, if that is something that you want to maintain or manage, it's good to kind of keep that upfront. If you were to hand a sibling a shopping list and extra, they could just go to the supermarket and just go crazy and buy whatever they want. But if you tell them you only have a budget of $50, go buy what we need for hamburgers, then they'll have a little bit more direction to kind of follow. That's kind of one analogy you can kind of think of it as.

Let's see here. So another question, "Does your agent automatically connect to After Effects or Blender? What does that process look like connecting your agent to those tools?" Yeah, you do have to have them downloaded manually. In terms of downloading the connectors, you can request Claude to go and find the connector for you, like scrape the internet and find the connector for the tool you'd like. In this case, I had already connected the After Effects and Blender, but when I say I connected it, that's as simple as me asking Claude, "Hey, does a Blender connector exist?" If it says yes, then I say, "Oh, go find it and install it, please." There's not much like manual setup other than actually downloading the software first. Definitely. Yeah, one thing that unlocked something in my head, because I know how intimidating and technical, something like what James just presented could look, but once you understand, it's literally just how you would conversate with someone that's actually going into After Effects or Blender doing this thing, just like 10 times faster and maybe more effectively.

It unlocks that for me. If you have a question, ask. If you need to communicate something, communicate it to the agent the same way you would do that with a person, honestly. That way, you can get straight to the point of what you need and nothing goes uncommunicated and things just go smoothly from there. If you've got a question, ask. There you go.

Let's see. There was another really good question about... If you are just starting, right now, I believe Opus 5.5 has the best usage value. You will be able to do the most tasks possible. If you spend the same hundred bucks on Claude as you do Codex, you will be able to do more things with Claude because it gives you more usage. Astra is a great model. Codex actually has a great UI and experience.

I started on Claude. That's why I stick with it. If you have never touched either of these surfaces, maybe ask your friends what they use. It's about who's the community of people you're around that you're going to figure this out with. There's tons of YouTube videos about people demoing. I feel like it comes down to personal preference if you're not at the level of technicality of "Oh, well, actually, Astra does this a lot better at 10 times usage." I'm just "I have all my contacts in Claude. I'm used to it.

I'm okay with the results that I get." Maybe when I evolve into a more technical person about this stuff, then I'll explore the nuances and differences between each model provider and whatnot. That's our answer to the Claude code versus co-work or GPT, et cetera. Let's see one more question. Go ahead. The Claude code versus co-work, use it in Claude. I'm sorry. That was the worst answer.

No, you're good. Use it in code because it works better there. It's really as simple as that. It doesn't need a long-winded explanation. If you're going to have the Claude desktop app, open it in code. Straightforward.

Let's see here. Let's see here. One more. I think that there was a question about printing. Since you're more knowledgeable about that, what do you think there? Does the FLORA 3D need to go to the printer's geometry slicer? The tricky thing about generating models with AI is you have no definition of the geometry and thickness.

I'd recommend using the MCP to manipulate model if you have traditional CAD skills, like taking the model in a software like Blender and prepping it for printing. You also need the printer's slicer because that turns your 3D model into what is called G-code. I'm getting a little too technical here, but pretty much to manufacture or print any object, it needs to be reprogrammed so it can run on the printer. Got it. Great. We'll pause questions there and we'll move on to the next section of the agenda. Let me go ahead and re-share my screen.

Was this Opus or Fable? Opus 5.5. Fable is really cool. It thinks for a really long time and I don't have the patience for that. I also have been working in Opus just as a default. Fable is being as powerful as it is. Sometimes you really don't need a machine gun to just get the job done.

Yeah. It's really good at designing decks and stuff. When it first came out, I used to use it directly in Cloud to make presentations. That was cool. We're past that. I don't even know if it's even that much better or different than the new Opus 5.5. Awesome.

Great. Next on our agenda is a live demonstration with MCP. You just saw James do some really amazing things with 3D printing, 3D modeling, all that stuff. Talking to James and seeing how he works, he mystified using MCP for me a lot. I gave it a shot on my own. We have some pre-built stuff that we can show you as well as we're going to live demonstrate how I would progress the project that I'm going to show you guys and go from there. James will be helping me out along the way too, maybe with some questions that might be running through my head and all that stuff.

It's chill. We're going to be showing you a real process of how this might look like with someone as non-technical as me. Let's go from there. I'm going to exit out of this and move over here. Direct the stone. Maybe or maybe not, you guys remembered. During one of our first floor labs, we had a wonderful time introducing people to the campus for the first time.

One of the images that I made was of a pancake, a stack of pancakes. I'm actually going to extend this out here. I had this image and I was it would be cool to carry the story of the stack of pancakes a little bit further than just a one-on-one. Off-generation. Me coming from a graphic design background, of course, I naturally lean more towards graphic design use cases for MCP. With that being said, I have some examples of turning this stack of pancakes into a pancake mix CPG product. I have examples where I'm taking this into social media ads, product photography, all that sort of stuff.

Then James is going to help me out in regards to making 3D assets out of it as well. Lots of different things that we can do there. That's the lore behind the stack of pancakes here. The story will carry on. As you can see, there's quite a bit on this canvas. One thing I want to focus on is the fact that the starting point for this canvas was literally just this image of the pancakes. If I shrink this back over, on my right-hand side, you see my Claude.

On my left-hand side, you see the floor of canvas. If I were to re-demonstrate how this would start out, my canvas would look like this. I open up a new canvas and I have a stack of pancakes. An alternative version of this might be, I have a brief, or I have a logo, or I have photography, I have product, I have a mood board, whatever it may be. In my case, it's a stack of pancakes. That's what we're going to run with. When you find yourself in a canvas like this and you're ready to experiment with how MCP can help you get off the ground, as James mentioned in one of his other demonstrations, up here at the top right corner, you're going to see Open In Agent.

If I actually click this, you're going to see an option for Claude, Chachi BT, Cursor, Grok, or General MCP. In this case, like we talked about, as you can see, I lean a little bit more towards Claude right now. There are two options here, Claude Code or Cowork. I went ahead and opened Claude Code as James suggested to me. I'm going to click this for now to show you what happens on the right-hand side. Again, on the left-hand side, I have my FLORA browser, on a Chrome, just browser.

Then on the right, I have my Claude. If I click Claude Code here, you're going to get that pop-up that James was talking about that open Claude. Yes, open Claude. On the left-hand side, this is what will happen when you click that button. You're going to get introduced to a blank chat. As you can see, I don't see it's called Claude Code, but you don't have to worry. You're not going to open up a terminal window with a black background and 8-bit text and all this language and code that you don't understand.

It's literally just a toggle on your Claude desktop app. Don't let the word "code" scare you. When you see Claude Code, just think, "Oh yeah, you mean Open Claude Desktop. That's it. Don't worry. Don't think too much about it." When I click "Open in Agent," go to Claude and click Claude Code. What happens in my actual desktop Claude app on the right is just this text populates in the chat area.

What I would do here is then press "Enter." What would end up happening is I'm going to move this down. Let me see here. Hold on. Let me move this over here. We'll go back to my chat. Is it this one? It's M3MCP sketch render?

No. Where is it? Here we go. Got it. Move this back here. Once you get that automated MCP connection text in the chat bubble, on the bottom, what ends up happening when you click "Enter" is a bunch of stuff like this. Don't worry too much about it.

I just let it do its thing until it asks me to actually do something. Whatever you see in this gray box is your response, obviously. Once I saw that it stopped running and it was "Tell me which one or something else and I'll give you the cost in USD before generating anything," it just finished installing or finished connecting to the canvas. It gave me two ways to continue. Use the empty space, like make an ad or a poster or build a series in the same style. I was "I have my own ideas. Let me communicate that to you." That's where this comes in.

I'm "Okay, great. As you can see, I have an image of a stack of pancakes in this canvas. I want to create an entire brand of pancake batter mix from this one image. I'm going to start with the following packaging concepts, product photography concepts, social ad campaign concepts, and for the visual direction, I want it to be minimal, but bright and vivid with good energy, but still sophisticated and friendly. In regards to color, I like the blue, but maybe for different flavors, they can adopt different colors.

Then for typography, I like bold bubbly letters that can lean pretty modern, but still readable. I also need help naming this thing, so let's start with naming it PankU. Use whatever image models you think is appropriate for this task. That's another thing. If you do want to dictate the image models or video models, et cetera, or 3D models, you can also tell Cloud to use those models specifically. Otherwise, you can just say, or you don't even have to mention it'll just go about selecting the models for you to the best of its knowledge. If it knows that chat GPT Image 2 is really good with text, and you're asking it to create a social design, like a social post with text, it will probably go ahead and choose that model based off of its background information about what it knows.

This here is my brief. What this might look like to you guys is, "Hey, Claude, I have this picture of a stack of pancakes. Here's a mood board of what I like in packaging design. Give me 10 concepts of what this packaging design could look like for this pancake mix based off this mood board.

Here's some inspiration for product photography. Give me 10 different concepts for product photography based off this mood board." That's another thing. I'm a big proponent for giving very high value reference images or reference whatever it may be, two agents or two AI models in general, so that you have to take less shots in the dark. In this case, I really wanted to just test out what it will give me just from a picture of a stack of pancakes. With that being said, I gave it this initial brief. Again, this whole everything from this point here is just Claude working on a few things. First of all, it came back with the order of how it's doing its thing, how it's going to section off places of its workspace.

If I'm being completely honest, I'm a little ignorant to this stuff. I'm just "I don't really care. Just show me stuff." If it tells me this whole process, I'm just "It sounds like what you're doing. Just let me see." I'm just "Go for it." One thing though, I want to make the whole graphic style pretty clean and graphic. It was done by a professional modern day designer. During one of James's points when he realized after seeing the generated video that you didn't want the white background anymore, I was "How many of us have been in that situation where you present an almost final asset and then you get direction of "Can we change that one thing? Can you just change it?" Can you go 10 steps backwards?

Can we just revisit that one little thing? It shouldn't take you that long. You don't know how involved it can be. Exactly what you were saying, a background change at this point means I have to go 10 steps back into Blender, re-render, re-light, re-whatever, re-apply the effect, re-import into After Effects, re-export, all that sort of stuff. That stuff naturally happens and MCP really works. It really helps tackle that pain point. You don't have to stress and panic the fact that you have a presentation in 10 minutes and your boss just asks you to do something that will actually take you three hours.

This was an example of "Oh, Claude already decided how it's going to tackle this thing." I forgot to mention, by the way, the graphic style should be clean and graphic. This was me maybe not as drastic of a change or drastically late in the timeline, but I was "Oh, one other thing though. Can you make the whole thing clean and graphic, et cetera?" It was "Great, I'll add that into the prompts. Let me just run it from there." It does this whole thing. It asks me if I want to approve the word mark first. As you can see here, it's saying start with the word mark, send an approval pop-up should now show in the Canvas tab, et cetera. I'm "Don't worry about it.

Just run. Just go." I'm going to go back to floor on my left-hand side and I'm just going to extend out here. Eventually, as I was asking it to run, and again, also one other thing because it was asked before, it does tell me the usage costs. Of these runs before it actually generates. That's another thing if you do want to be cautious of that. It will just show you the cost of it before actually using that. I'm just "Cool, run." It does its thing for a little while.

It gives me a few little previews here. As you can see, a few of them apparently fail, but as you can see on the left-hand side, I don't see any failed generations on the Canvas. You can see it troubleshooting. I don't see any failed generations on the Canvas, but I don't see any failed generations on the Canvas. I don't see any failed generations on the Canvas. I don't see any failed generations on the Canvas. I don't see any failed generations on the Canvas.

I don't see any failed generations on the Canvas. I don't see any failed generations on the Canvas. The word mark is different than the word mark that it was producing for me on the left. I know it's small there, but it recognized that. 11 of them hold the brand together. I think over here, range lineup is off brief. It came out as a generic milk carton with a plain Panku logo, not your word mark or the box designs. Nano Banana Pro lost the design here even though the four boxes were attached as references.

It went wrong, and you're recognizing that. I love accountability. Very accountable. It's almost like this is a tragedy. I can't believe I'm even showing you. I'm so sorry. I hate to tell you this.

That's a good one. Oh my God. But yeah, I was kind of "Okay, Claude, I'm trusting you with this process now." I'm just "Okay, it's fine. Can you try to fix a generic milk carton logo? I'm not in here like micromanaging Claude. It probably knows more about design than me at this point, to a point, to a point." I just kind of let it go. As you can see, it went ahead and corrected the generation.

If I kind of go back to my code, I have a canvas on the left-hand side, and I zoom out here. This was the first generation where it was kind of freaking out how bad it was, and then this is the new generation after I told it, "Just try again until you get it right," and it got it right. There you go. That's one case where Claude says, "I made an oopsie," and you're "It's okay. Try again." At this point, I'm "Okay, great. Can you take the nine by..." up until this point, before we get there, if I look at this workflow altogether on the left-hand sidebar, and I'm going to go ahead and extend this out. Again, I started this canvas off like this, and I got this.

I just want to emphasize something though. I'm completely aware of the fact that almost no branding project or almost no design project starts off with "Just take this image and make a branding project out of it," or "Let's take an image of a pair, a stack of pancakes, and imagine what a product out of this would look like." I am completely aware that projects don't usually start off like that, but I think the point of this demonstration is you can start off with a simple of one image and blow this out into the foundation of a system that you can feel confident about building off of. Now, imagine if you actually gave the agent a brief, an existing name for the brand, an existing brand story, photography, brand assets, or mood boards for all these different layers of this project. The possibilities are endless. I know this seems to more high-grade designers that this is a very surface-level example, but I do want to emphasize it's just a point of if you can get this far with one image, imagine how far you can go with actual assets for a project. If we can just take a look at what's on the canvas that we got from just this pancake image, we got a starter basic parent core product of some pancake mix, then a blue box, and then we got a blueberry mix, we got a chocolate chip mix, and then a banana mix. We also got a few nice little product photography shots that are a little bit more involved than just the box.

I love this one. I'm going to actually point that out. I'm going to mark this red. Mark it and just say, "Hey, do you see that thing I marked red?" I'm going to do that right now.

Let's see that. We're going to be like- This is not chill. Okay, let's see what happens here. I'm actually going to go in here and say, "Hey, I noticed that there is-" Actually, I would start this one by saying, "Do you see the asset I just tagged in red?" Okay.

Then say, "I noticed." There we go. Do you see the asset I tagged? I realized that the box design is for the classic product, but there are blueberries in the scene. I think you meant to put the blueberry mix in this image. Can you try again? Variations with different color backgrounds.

Let's go from there. Let me exit this. Let's see how it works live. I'm actually going to zoom out here so we can see it work. I'm going to zoom out here. Let it do its thing. That's a really good point.

Sometimes, I didn't even catch it. Maybe in my head, it was just maybe you like playing pancakes with blueberries on the side. I see the moment I was critiquing it. I was "Wait, maybe this can just be fun." Maybe Claude just likes blueberries on the side. I don't know. Sometimes, I just like the pressure, Claude. Bro, we got to try harder.

I try to be as nice as I can to my agents because I watch too many movies. It's like collecting karma, collecting digital karma points, but that's besides the point. That's a totally different topic. It's a really good point. That the MCP did on the canvas that you want to be "Hey, I think you made a mistake here." You could just go back into it. James suggested label it red and be "Okay, first of all, just reinstate to the agent, do you see the asset I tagged in red?" It'll know immediately "Oh, there is an asset I need to pay attention to specifically. Let me go pay attention to that in context to whatever else I'm writing here." As you can see on the left, those nodes just popped up.

I'll move over here so we can see that. It'll go ahead and just keep doing its thing. I think it's going to ask me for usage or it's going to go ahead and run it. >> Well, tell me all five of those are only 90 cents, which is dope. You said all five are generating all the review when they're done. >> Nice. Those are going to generate and then once those are done, it'll get back to us. It's a really great way to also build a design report. One of the most frustrating things about being a designer sometimes is you'll spend so much time and effort on a particularly designed asset.

All it takes is for your art director to look at it for one second and be "There's a typo there," or "That doesn't belong there," or "That's the wrong logo." How did I not see it? You feel a little insecure. You're "Damn it. How did I not catch that?" I realize it's because it's easier to critique. It's easier to look at something and pick out what's wrong with it than it is to pick out what's wrong with something you made and spent the last few hours making. Do you feel the same way? >> Yeah, because you're focused on very different details. I'm thinking a graphic designer and you're focused on, "Is it going to print properly?" >> Exactly. >> Meanwhile, you got 50 blueberries and you're trying to pancake mix it. >> I know.

All it takes is one extra fresh pair of eyes to look at it and just be like. >> Now, you're your own fresh pair. >> Exactly. Okay. Whoa. Unlock there. MCP makes it really easy for you to look at something with fresh eyes and approach it from a directive or a collaborative angle. Because naturally, in the creative process, it is so much easier to look at something and find something wrong than it is to do that yourself. Of course, some of the best designers will be able to look at their design after five seconds of looking at a white wall and be "I did this wrong and that margin isn't right and that A is different than that A," all that stuff.

This is also valuable in its own way. I'm in that regard. Let's revisit this here. Let's look. One thing I noticed is that it also changed the composition a little bit, which I know why because this composition is not plugged into those new assets. It's re-interpreting that composition, but I'm actually not mad about it. >> No, I am mad about that. It's like you're not using the references properly.

It's "Dude, get better." >> Okay. >> A couple of things here. >> Yeah. >> When you're in Claude, it has a tendency to not want to pick chat GPT. For this specific use case, because text is critical, I would tell the model like going forward, only use chat GPT and that's like using that as part of the reason we're failing. You also aren't using the reference inputs properly to retain the layout that we defined. >> Amazing. I think I'm actually going to do exactly that. A few things. One, since these assets are text heavy, I'm going to pretend like I'm ignorant to the fact that chat GPT is good at text because maybe coming from someone that doesn't know everything about models, I'm actually just going to be "Since these assets are text heavy, can you ensure you're using a model meant for fidelity?" >> I'm curious what it says here. >> Okay. >> I'm almost positive that it has a bias against using chat GPT, but let's find out from this question. >> Okay. >> If you ask it not hard on it, I would saying "Dude, get better." >> Okay.

Let's see what it says. That's so funny. >> Yeah. >> Okay. >> It's perfect. Don't ask it like I will. >> Okay. >> Yeah. Let's see if I'm correct. >> All right.

Let's see what Claude is about here. >> Let's see what it's about. >> It's going to be "Actually, Nanobanana is totally fine." >> I'm going to be "I don't know about that, Claude." >> Yeah. >> "Find me resources on the Internet why Nanobanana is better at text fidelity." Let's actually put this to the test. Okay. A few things. Since these assets are text heavy, can you ensure you're using a model meant for text fidelity?

Then two is I notice that the compositions have changed from the original image reference. >> But I believe if we go back and look only that one, the red one, it was a first iteration of the composition. >> Okay. >> So you need to tell it like I want to lock the composition, so use that as a reference to fix the blueberry incident. >> Got it. Okay. >> Because we're really trying to solve a simple problem. >> For sure. Okay. >> But we were so creatively inspired we chose to iterate, which is there's nothing wrong with that. >> I think that makes sense. Let me reword this then. You're saying that I need to identify to the agent. >> There was a locked composition from that generation I was originally referencing. I just wanted to change the box and the background. >> Yes. I would say this.

The image I had tagged in red had blueberries I didn't want in it, but it was the layout. >> It was a good layout, but the composition was fine. >> The composition was something I wanted to carry over. >> Was something I wanted to carry over. >> With that said, you need to use that as a reference to generate the output that I'd like. Does that make sense? Ask it, does that make sense? Sometimes I'm yeah, am I crazy? >> One thing I also want to mention, how coding language in general, you're literally learning a new language like coding in general, like you need to do like brackets and like if XYZ then ABC one, two, three. I don't know like coding in general is just like its own language. And so it's like really intimidating.

But when you're talking to an agent, it's literally just you practicing how to communicate something very clearly. So if anything, maybe kind of communicating to agents is also making this better human communicators. >> Definitely. And also I'd never use the terminal in my life. I don't know how to do anything other than press enter. And every time I've ever used a terminal, I've had chat LGBT walk me through how to use this terminal. >> Yeah, no, get me away from terminal. The second I see like a black screen with like green letters, I'm I've checked out. >> Yeah, tune me out. >> Too funny.

Let's see, I'm going to let that run. Okay, actually, yes, that makes sense. The host of tools don't show color tags though. So the host of tools don't show color tags.

So maybe you can't see the red. >> Okay. So you're telling me you can't see this as tagged and ready. >> Should I stop? >> No, the best part about being in code is that you can give it like it will think about the next task. It can use it's using sub agents. Once you do something, it's ready for you to ask another thing. And like it can juggle executing an operation and thinking about your next one. >> Got it.

Okay. So wait, let me see here. So yes, that makes sense. >> Okay, so first of all, let's fix that and use N14 itself as a reference image to preserve. So N14. >> Is that like the name of the generation on the canvas? >> Yeah, definitely it is. >> It definitely is.

Okay. But I'm curious if I think it might actually be working off of this. >> So this is where I would like exit out of answering the question. >> Okay. >> Use this screenshot and say, so you're telling me you can't see the red mark I tagged in this thing, not this thing. I know I tagged this image with because it definitely should. But sometimes I do think that it's based on its own language, right? Like N14, like I don't ever want to know what number. >> Yeah, I don't. >> Because it's thinking with chronological order. At this point, I've already forgotten what order we're in. >> Yeah, that's a very good point.

Okay, so let's see. I can't see it. N14 photo flat lay. >> Say going forward, never tell me the chronological order of a node, communicate the nodes to me by the name of them. >> I'm going to be nice, please. >> I don't have time to be nice with it. >> I always, I got to have time to be nice. >> Go with the fade karma. >> Karma is real. >> Be careful. >> Yeah, be careful people. >> Actually, talk to your AI very nice. >> For real. It's good practice for real life. >> No, it is. >> It carries over. Definitely like even AIs treat like a human. Be more respectful. >> There you go. >> I'm going to do that going. >> Yeah, it's good practice for the real world. We have to remember that we're all humans that we talk to each other. >> That is a clear state. >> Anyway, there's your little piece of woodwind. >> Okay, so we're going to be nice to it. >> So going forward, please don't refer to nodes in their chronological name. >> Yeah, don't refer to nodes based on their chronological order. >> Order. >> I have no at this point in my creative process.

I don't know what the order of them are and it's not helpful. >> No, no, that's valid. That's valid. Okay. So order they are in and they're. >> That name and convention. >> Information is too difficult. >> Isn't helpful. >> Isn't helpful. Isn't helpful. >> Just say that. >> Okay. >> There's no way, can't read them. >> Interesting.

So we're going to pivot here. >> We're going to do a pivot here. Actually, if it's saying it can't read the node, I would change the name of the output up here and call it a red photo flat way. Then copy and paste that. >> Oops. >> Do shift. >> Yeah. >> Yeah, and say, can you read nodes by the titles we give them? I just changed a node's name to red photo flat way. Can you see that adjustment?

Let's confirm what you can and can't know because that's really weird that it didn't read through it. >> Awesome. Okay. >> But also click browser. >> This one. >> I didn't know this. >> So import cookies. >> I didn't know this. >> This is why it couldn't read it needs to be open. >> Do I have to import? >> Yeah. >> Import? >> It wouldn't hurt. >> Okay. >> Because this way, we can hold off on this. >> Okay. I'm going to go back. >> We're not doing this. >> Okay. >> Going forward, the reason it couldn't read the node was because the Cloud Code didn't have full access to our Chrome browser and that's the block. >> I see. Okay.

So do you recommend me leaving this open instead of just having the canvas or the- >> Yeah. I would do it like this. >> Got it. Okay. >> Just whenever you want to go to a full browser experience, just reopen your Chrome tab. >> Got it. >> You can place things in here. You can change the name of anything. You can interact with it just through here. >> You see guys, I may be enough to do this, but I didn't actually know there was even a browser option. I was preparing for this webinar. I was I got to have my Chrome next on the left-hand sidebar and my Cloud on the right-hand sidebar.

James is just here "Yo, there's a button for that." Just literally click the browser. >> Yeah. Now we're in full web MCP. >> We are in full MCP mode. I am fully engulfed now. >> Engulfed. >> Okay. >> Let's go back here. So got it. From now on, I'll use no titles.

Okay. I can see your rename. Okay. That makes sense. >> Cool. >> All right. For the new versions, each one will be an edit of red photo, flat lay and chat GPT Image to keep in its exact layout, et cetera. >> Oh. >> Yeah. >> I was just going to say, do you see how it's already provided you like a suggested, like you saying go forward. If you click the chat box and press tab. >> What's it's going to do? >> It's just going to fill it in. >> Whoa. >> Then you just press enter.

So once you're deep in a project, at this point, yeah, it's just like tab. We know what we're doing. Let's keep pushing. I don't recommend you have an idea and you're one step into it. >> Yeah. >> You're just like tab suggested response. >> Right. That's not as valuable. You don't get as much out of it.

Okay. That makes sense. Interesting. >> Click off the comment. >> Got it. >> Yeah. >> Okay. There we go. Damn.

Okay. All right. Let's let that go. But yeah, that is super interesting. I was excited to do this webinar with you because as I'm too busy to dig in MCP, but now I'm ready. >> Let's go. >> I'm ready. I'm ready to download Blender. I'm ready to learn how to use a 3D printer. >> This is a big moment for us. >> Yeah.

It's another way I like to explain MCP connections with any of your creative tools. I'll actually show you something else after making this point, but I think they're just generating. >> It's generating sandbox. >> You can't see it through the live. >> Yeah. But it's just going to like in a second, it'll spit it back out. >> All good. >> The point that you can think of MCP as is as long as I've been a designer, I feel like I can only actually use Adobe Illustrator to 30-40 percent of its full capacity. There are still probably 90 percent of the tools in Illustrator I don't know how to use in the toolbar. I literally just stick to the pen tool, the shape tool, and six other things. If I were to use MCP via Illustrator or any other creative tool, I feel like the MCP will know how to get the most out of those tools because it just technically knows the backend of what the tools are meant to do. It's accessing information way faster and way more effectively than I could ever do.

Plus, we're all working adults here. Yes, could someone make an argument of hey, just learn the tool, just do the thing. It's okay, but I work eight hours a day as a creative, and then maybe if I have to make things up, I have to do some freelance after work. Then after that, I'm a human, I would like to decompress. You're telling me I need to spend another three hours every night learning every nook and cranny of a creative software. It could be a lot. Of course, in a perfect world, we'll all have the resources to have that luxury, but with that being said, if MCP helps you get the most out of the software you're already paying for, why not do it?

Because you are able to do a lot more work. You're able to confidently get what you want out of the software with just a little bit of help from an agent. After we take a look at these, I'll show you how I actually use MCP to then take one of these designs that were generated in FLORA into an editable file through Illustrator. Let's take a look at how Claude went about correcting its initial mistake. There we go.

Now, we have the blueberry pancake mix instead of the regular pancake mix, but my blueberries are gone. This is where it's yeah, we got to stop being nice, bro. You got to do better. Come on, if my junior designer came back to me and stripped the blueberries on to put the wrong packaging in Claude's defense, you said no berries. And you did. Bro, you told me to press tab. Oh, okay.

This is what happens when you get tab happy. Thinking it knows the right answer. It said, do you want the blueberry box and no blueberries? It's die. Why would I ever want that? Put the blueberries back in, please. The blueberries back, dude.

Goodness gracious. Sorry about that. Sorry, I should have read your response or your efforts. Thank you for your efforts. No, not thank you for your efforts. Well, I mean, I gave you bad. I gave you bad direction and that's on me.

Sometimes I have to confirm dude, I messed up big. Claude, I'm so sorry. I made a huge mistake. I made a big one. Oh my God, that's too funny.

Okay, great. So let's let that run for a bit. And wait, I saw you earlier kind of like highlight these and run them manually. Oh yeah. I was just I've trained my Claude account to always generate things live. Sometimes MCPs, you make a request, it will do the request in a what's called sandboxed environment, meaning it's doing it in a cloud. We can't see some nerd stuff.

I don't need to know. And I don't really care about it as long as I'm getting what I want. If James doesn't want you to know about it, I definitely do not need to know. So sandboxes, if you don't know what they are, it's all good. Like you don't really need to know what they are. Sandboxes decide the moment where we saw the text node with the prompt in it, and then we didn't see it run.

And then it just spawned the output because it's sending it from somewhere else. So what I did in the meantime, because I just wanted to see them run, I selected them all and I pressed the play button and then it ran all five of them. Got it. And again, like you do see the usage here.

So if you did kind of want to just like pay attention to that before you click the run button, that's also an option. But while that's running, let me show you how I kind of went about taking one of these designs and bringing it into, or turning it into an actual like editable file that I can then manipulate. So I think I'm going to go ahead and scroll all the way back up here. Okay, cool.

So I think I could zoom in too. Well, there you go. Now we're both learning. So I think at this point, like after I had corrected the off-brand kind of milk cartons, I was okay, great. You can take the nine by 16 social posts. And again, I know we just went over like labeling the node and like renaming it and naming it or like calling it out by its name.

But here, this particular generation here was the one that I kind of want to recreate in a vector editing tool. So I didn't actually personally know the name of the node. I know it's up there on the upper left hand corner, but that's besides the point. So it was just all right, great. I went into cloud. It was can you take the nine by 16 social posts and recreate it in my Adobe Illustrator program, split it into a layer for the type, a layer for the pancake and crumbs, a layer for the pancake box, a layer for the background and a layer for the button. A few things here.

I had previously already connected my Adobe programs to my cloud. But what exactly does that mean? Because that's like vague in itself. You can literally just go into a fresh cloud chat and be or even in this chat itself, I just be I need you to get access to my Adobe programs. Can you just what do you have to do to get to make that happen? And leave it at that. Cloud will kind of do the thing that needs to be done and then eventually it'll tell you all right, let's I'm ready to go.

What do you need me to do in this program? And it'll kind of go from there. And then that goes for that goes for Blender. That goes for any creative software that you may have. You name it'll do it. If it can't, it'll come back to tell you.

And then you can yell at it a little bit and it'll do a better job. Or you could be really nice and it'll do an amazing job. So from there, yeah, that's that was kind of my prompt initially to kind of just be all right, let's now turn this into something that I can actually edit. So I went ahead and just entered that. And, again, I'm not going to read through this. I don't read it on my own time. I will read it to you guys.

It'll be a bedtime story if I do just kind of let it do its thing. And then eventually it tells me no, eventually that's when we started yelling at it about the tagged red thing. But it's telling me eventually OK, these are the layers I made. This is what's on the layer. And is it editable?

So if the type layer is editable, it'll say fully edit the text, font size, et cetera. So if I actually go here. And I have this here on the left right hand side bar. So this was generated on the right hand side bar in FLORA. And this here is now an editable design. And I was curious about this, too.

So things to know. Headline font, it's set in SF Pro rounded black. The closest installed font. It literally was hmm, what's the closest installed font to like the rounded kind of bubbly font you have in the poster? It's a bit less inflated than the lettering in the generated ad. Apple's license generally doesn't cover SF Pro and marketing work.

But we've also picked it for your ABC publisher rounded mix is only a trial version. So I have like a bunch of downloaded one of my past times is like going on type foundry websites because this is all I ever did at my job at Pentagram. Like just go on type foundry companies and download as many trial fonts as you can and just like look at them. So I have tons of them installed on the computer. And because they're not purchased at their trial fonts.

So I only use them if like I need to buy by the license, et cetera, if I'm like exploring. But it actually like went ahead and was that ABC publisher rounded mix you have that's only a trial version, you could probably like try to see what that might look like since the text is editable. So I'm okay, type designer Claude, like I see where you're coming from. So yeah, it went ahead and just picked the font that it thought would fit in this case.

So hello. I'm an editable design now that was originally generated in flora, which is amazing. And this pancake and crumbs are these pancake and crumbs are I didn't know you could even mask stuff. I've used illustrator for a long time. I don't know that you can mask things.

This is actually not masks though. What is it? This is just a transparent image. So there are no vector lines that are masking this. It literally like took that part of the image and like isolated it on a trend as a PNG on its own. It split the layer. That shit is crazy.

That shit's crazy. Yeah, since it's editable, like now what started off is a pretty standard like design layout of oh, type at the top pancake in the middle and box at the bottom. It's all right, now let's let's remix this a little bit. So like now I can copy and paste this. I'm let's see what it looks like if the pancake mix was like smaller. What if the what if the button here was kind of moved over here? What if it was no box?

What if it was just the pancake? All that sort of stuff. So all of a sudden, the point is you are no longer locked into just the capabilities of one program. And we really want to one thing we really emphasize that FLORA is our tool is supposed to promote creativity in general via generation generated, like diffusion models that we have in the platform, and the way that you can connect them to kind of make your own AI model cocktails. It's it just makes it way more naturally kind of it allows for a more explorative environment and a more creative environment in a canvas like FLORA. And so when you get to something, of course, it's a little bit more of a generic layout, okay, I get it.

But if this was like off the chain, damn, that's actually a really good like starting point. I see that and like 100 different images kind of pop up in my head. You have a choice. You can then generate those 100 images in FLORA or you're I want to take the step in and now mess around with the type mess around with the sizing, all that sort of stuff.

But the pain point of that process is, shit, I need to rebuild this in Illustrator. Not anymore. That pain point is gone. But the creativity is still there where you choose to have it. If you enjoy rebuilding things for if you chose here, I was like happy to kind of just like take this into Illustrator, mask out the crumbs, mask out the box, relay out the type, all that sort of stuff. That's a beauty of choice, baby. Like you can go ahead and do that.

But if that is a pain point, that pain point can no is no longer there to kind of stress you out and like burn you out through this creative process, that is kind of supposed to be fun in the first place. So personally, I like to use MCP and agents to hand off the busy work, get me closer to the fun part. So that I can spend my creative energy doing the stuff I actually enjoy. So with that being said, something as useful as kind of taking a generated layout and bringing it using an MCP to just kind of rebuild it in an editable software that you're comfortable with is extremely powerful.

So definitely something I like to kind of emphasize than here at FLORA, we recognize that and we kind of want to just promote as much of a creative process for our users as possible. So with that being said, let's go down here and revisit how it did with adding in the blueberries. Okay, wonderful. There we go. It did its thing. Actually, I kind of want to see if it responded. No need to apologize.

Yeah, you're right. There is no need to apologize. Let's keep pushing. I'm a natural people pleaser.

So I just naturally kind of gravitate towards that type of language. So making friends, I know. This is me. It's oh, well, what do we do? It cowers at the corner. Oh, my God.

But yeah, it's a no need to apologize. I'll add the blueberries back as new versions, etc. Okay. Great.

So as we can see on the right hand side bar, the blueberries are now back in the composition. And I'm, I'm pretty happy with it. I kind of like the monochromatic kind of effect going on here. It's kind of cool. Oh, yeah, go on. It'd be cool if we told Claude that make an animation where all of these products like slide in let's do it. We got time.

We got time. All right. Let's so now we've been sick in a lot of this whole thing has been like very static images. This one hits.

This one hits. You like the color. I like the contrast. No, no, let's go with that.

Let's go with that. Cause that makes sense. I think like purple and yellow are like complimentary or like contrasting colors. Let's go with it.

Okay. So actually, do you want to go ahead and drive this? Yeah. All right.

Let's go. James is in charge now. Okay. So I'm going to go just command C on this image. Oh, you have to copy the name of it. No, honestly, I just like to screenshot stuff. Yeah, go ahead.

I'm a big screenwriter. On the bottom here. Hold on. I got you. I got you. I got you. Hold up.

There you go. Let's try this again. Destroy it. I got it. I got it.

Okay. Screenshot it. Yeah. What a team. Dream team.

Okay. Our two brain cells combined. Got it. Okay. Boom. All right. There you go.

Okay. So I, one, two, go on each element. Oh, there you go. That's a great element. I needed that word. Each element, is pushed. Try to do it.

As if like you didn't have fully formed thoughts, you were like word vomit is pushed into a frame by hand. Okay. I didn't know by hand, one by one, you will need to me. Any empty first frame. You'll need to make an empty first frame. What would happen if you didn't give it that direction though? Like there'd be some hallucination at the beginning of this video where like the things were there.

I see. I see. Okay. So that's like a best practice thing. Yeah.

Okay. You will need to make a first and empty first frame for this to run once you generate the second. You're permitting. You want it six seconds.

Okay. All right. Use. I'm not seeing ads for, image. Oh, for video, there you go. Right. Yeah.

See that's 2.5. Oh, and like. This is what I was talking about with Claude can understand like multiple thoughts at once. It was already acting on the first one.

And then I was Oh wait, I left the detail out and then it's going to add the detail. Because it can think and process things so quickly. There you go. And it let me know. Oh, I caught that. I'll switch. Nice.

All right. So let's wait for that to kind of go boom. So first thing is doing is making the empty first frame. Damn. Yeah. I know. I, one of the things like regardless of whether I'm using an agent or like I'm in the canvas, I'm always just using text notes to help me write prompts.

It's one of the first things I like naturally learned here. Just use a text node because I can't prompt like James can, and I can't prompt the way that Claude can. So I don't think I ever have to, why do I have to fight it? Okay.

So let's see. Yeah. That one's a good one. Our blueberry incident was unacceptable. It looks like it's running them on the sandbox. I wish I always did it live. We'll fix that offline.

Okay. No problem. So I'll leave it as is, or should I go ahead and click these on the side and run? For the sake of time, like click it just in case it takes long either way, like the spawn. Very cool. And in this case, it's making the image frame before it uses both frames to make the image.

So we can't run the video yet. Oh, right. Because in general, it's following like workflow best practices and like when you want to generate a video, best practices to always start off with as close of you of a reference image as you can. Got it.

Okay. That makes sense. All right, cool. So while that's going, let's take a look at a few questions here.

Let's see. Okay. So I see foundational question. I still have to enable FLORA MCP on Claude to get started. Or from FLORA.

So would you recommend connecting the MCP while you're in a canvas? Like with that at open agent button, or do you recommend opening Claude first and then connecting FLORA from Claude? I think just going into the canvas and clicking the open agent button up here is pretty straightforward. Yeah, I agree. I think that's the move and it's the easiest way. It, when you say foundational, I'm assuming like you might not have the cloud desktop app.

So, I'm actually releasing a video. This week about setting that up. The first time you open Claude code, it's going to ask you to install on your computer, get it's clicking one button, but you need to click that. Let it like load the software. It takes like 20 minutes to download the software on your computer. And once you've downloaded the get software, then open a FLORA canvas you're working on or a fresh one and use the open an agent button. That's just like the prerequisite.

Nice. Okay. Let's see here. Ah, wondering, I still need blender and After Effects licenses to make all of this work. Or can I refer to there and no, you, the skills and MCP call-ons only function. If you have the tools, when you talk about licensing's, blenders free. Oh, After Effects has a bunch of open source competitors.

So I'm assuming that the, this question's coming from the mindset of Oh, I love using a bunch of tools, but I don't want to pay for 10 tools. And I agree. So I try and find like open source free tools like blender and there's plenty alternatives to After Effects. Yeah. Remotion is my favorite. Definitely. That's enough.

Yeah. To that point, like one thing we, one thing we do want to emphasize is we're not telling you look, we can have this like conversation about how expensive like creative, software is even before like generative usage came into the picture. That's another topic that's maybe like a fireside chat, like hot mic type moment, but, yeah, exactly. To James's point, if you do find like open source versions of like video editing software, there are plenty out there. And the thing is you don't have to feel like you'll have to learn them all over again, because to my point earlier, the agent will kind of be able to decode the new nuances of even an open source program, and kind of get the most out of it from there.

So, if you ever seen an MSN like people in MSN paint, MSN, I mean like Microsoft paint painting like the fricking Mona Lisa, like you can work with something as simple as that type of art program and create some really beautiful pixel art. It all just depends on, what you're trying to use that software for. So with that being said, please don't let like lice expensive licenses for creative software kind of stop you from creatively exploring. At this point it's the door has been blown open. Please like explore to your creative desires in that regard. And yeah, you can kind of go from there. I couldn't agree more.

You can find free versions of illustrator and Photoshop on the internet. You could also make any creative tool you want with cloud code and like put your own tool on the internet. There's really like no limitations to creative tooling in this era. There you go.

Let's see here. Let's see here. Okay. So not a question, about MCP specifically, but I work in CPG and we use FLORA a lot. What would you say is the best model to keep the text sharp? I want to use MCP more often, but when it comes to packaging, I need precision.

So when you use using, when creating like very defined, like design assets that use text that use like logos where things have to be pixel perfect, because again, that's like very important and something like CPG. One thing I do want to emphasize is like generative models output, flat images up to a specific resolution. So no matter what, unless there are SVG models out there, but in regards to like what we're used to for the last like three decades of how like traditional creative softwares work, where things are layered, you have a raster layer and then you have a vector layer of the logo right on top of it. What you get out of an output, from an AI model or from like a diffusion model is flat.

So a really good example is the thing that I did here where I generated this social media, kind of design. And of course, this text looks pretty sharp there. Allow for the session. Yeah, right.

Okay. The text here looks pretty sharp, but again, I'm sure if I looked at it with like a microscopic, like magnifying glass as a typographer, you could probably find, Oh, the anchor points in this A are different than the A in Sunday or something like that. But with that being said, like this is probably 90, 99% of the way they are in regards to how this composition should look. So then that's when I would use like an MCP to kind of then okay, I like where this is at right now. I need to make sure that if I care about typography, I want to do that myself.

So I can guarantee that if someone asks me how the typography was added to this composition and maybe instead of make Tuesday feel like tastes like a Sunday, it's going to be make Monday tastes like a Sunday. So then you can just kind of go back in here into the, layerized version. Of that composition in your, editing software and kind of just adjust it from there. But the fact that like you can go from a, something that is generated 90% of the way there in regards to accuracy and then turn it into something that's what you would define as a hundred percent accurate, via MCP connection and whatnot. That's also, kind of one way to go about it. Because yeah, I know text is kind of, really important in that regard, but just a simple answer to best model for keeping text sharp chat, cheap, GPT Image models do really well with generating very accurate text, kind of representations.

So GPT Image models will be my answer to that regard. When in doubt, just use one of the newest models. I know they come out all the time. When in doubt, just use the newest model or the newest version and kind of go from there.

Okay. Let's see if this video finished or not. Okay. Still going.

So I'll leave, I'll leave that there. Kind of go through a few other questions here. Okay. So Sue me asked in this particular case, if you have specific brand assets, like a logo images or references, would you place them on the canvas like James did or attach them in your brief to Claude? I prefer just placing them on the canvas. If you're, like already proficient in Claude and how to like connect your local folders. Yeah.

I just, I really find a lot of value in placing all the assets on the canvas and then selecting open an agent. And then Claude's first response is like a breakdown of everything we're about to work with. Like I haven't really defined like how powerful that is or why I feel like it's so powerful. It's just I would recommend that. Awesome. Yeah. One of the things, first things that James kind of told me to do.

Was start off by organizing and it's, it's actually one of the best practices points that we'll kind of finish off on. Putting all your assets into canvas that you're going to hand off to an agent with like clear, labels of this is a mood board for logo design. This is a mood board for color. This is a mood board for this, or this is a logo for this. And here are like some extra details about that's also one way to go about it.

So yeah, kind of experiment with both and see what kind of fits there. Let's see here. Okay. So question for James from Kim, as I think Kat has told us before in another session, what models do you use for different use cases and tasks? You mentioned chat GPT for text.

So I guess this is just more of a general kind of question in regards to model models you prefer for different things. I number one, my favorite strategy BT. I Korea for like art direction and taking a bunch of references. Off Pinterest for like an aesthetic. I'm trying to recreate and using Korea for that. I also like nano banana for taking two different images that if they're not text heavy and composing them together, I think that nano bananas, like high level strong suit is composition of things. Good point.

Yeah. I feel like that was a really good all around her. Yeah. Okay. All right.

Let's take a look at this video and see. How this came out. Okay. Wait a minute. Oh, let's go. Or the finger. Do we have five fingers?

Okay. That, okay. Good. I was just checking. I always like to damn, that did a really good job. I'm really, the lighting is really nice. I'm not going to actually send this because why use, I'm not going to, I was going to do something funny, but I was going to say oh, I'm actually really proud of this.

Yeah. No, that's the one thing I do at the end, because at the end of it, I'm what, bro? Like I'm actually pretty proud of you. Does that, okay.

So honest question, does it actually make a difference on how it will kind of like learn to do things for you for future reference? So, let's when I made the, the video about like the synopsis about everything. And, if I liked how it made the video and I felt like something, maybe, the way it placed the node connections was significantly better. I will say, I actually really how much, stronger our node connections look. They feel more on brand with the flora, identity, go ahead and make sure you update our skill that we use for this task.

So I like to, give it positive reinforcement when I want to add to my skills. Very good point. So I know we didn't know. We didn't go over skills too much. And for the sake of time, we'll kind of see that for another, floor lab, but I was about to be Oh, I actually don't want to like use the usage just to kind of I'm proud of you and just do that.

But if I'm kind of using a skill that I wanted to update based off of the things that I did right this time, then it's actually like valuable. Cause think about this. Like if going forward, you had, you got the reverse hover. Yeah, I know. I have a windows user. I was a windows user. I update all my max okay.

So if ever again in your life, you thought you were going to start with just a product and not know how to brand it, you could make this a first experience, a like CPG branding skill and say, take everything we learned from this. So I could apply it to other products and call it the CPG flora skill. Let's see if we can actually share this skill. She's still to shareable, right? Yeah, we can post this. There we go. You guys are going to leave with a.

Not leave stay tuned. Before I promise the world you're on a silver flatter. Okay. So take everything we learned from this and make me a skill called CPG flora skill.

Okay, great. Awesome. So let's kind of see where that goes. I know we're going to come up on time soon and I want to leave a little bit of room for a few closing remarks.

So let me go ahead and switch back to my presentation over here. So, first of all, if you're here at this point, thank you so much for tuning in. I know that there are a lot of different things about our like educational content that we're experimenting with. Some of them are three hours. Some of them are two. Some of them are one. Let us know, do you like longer format kind of like live sessions like this?

Do you prefer kind of like asynchronous, kind of learning, all that sort of stuff is like really valuable for us so we can kind of continue to give you guys the best educational content, that we can. So, I know also this was in the middle of the day. We kind of wanted to be cognizant of our, international, users. But first of all, thank you guys so much for joining us today.

This was a lot of fun. Me and James, we are so heads down sometimes at work. That and he sits right behind me and sometimes I'll be yo, what's up? And he'll be yo, and then that'll be our interaction for the day.

But this was honestly like a lot. That was like our interaction leading up to this. I know is are we going to do this or what? Of course we will come on.

So I'll close out with a few best practices here. So we went over way more than this, but, best practices in regards to using MCP, you get the most out of an agent when you give it something to work with. If your creative brief for an agent can be a rough workflow of what you want. Another thing is use labels on a board to organize what assets or what workflows should be used for.

And then finally, give a task to the agent and ask what the agent needs to run a successful workflow. Cause that's also valuable. Sometimes you can add, have a particular briefer, requests and the agent can be Hey, if you want me to do a good job, I need a few things. So. Closing out again. Yeah, we host these every week. You can find more upcoming webinars via our Luma link.

We will be following up with an email to all attendees with a one month free floor of pro. And this recording will be available to rewatch on our YouTube after some post-production editing. So with that being said, thank you guys so much for, kind of attending and Olivia, I'll let you close it out from here. Guys, an amazing session. Unsurprisingly, Kat and James, great work together. What a duo we have here. Yeah, I think we covered it.

The cat, this session will be sent out and again, we will be sending out also a code to try out FLORA. If you are a new user, we'll be sending out a code for a free month of FLORA pro. And yeah, like Kat said, lots more to come on the education front. So I sent the Luma link in the chat for our full education calendar.

So do take a look. And see you guys next time. Thanks for coming. Thank you guys. Have a good one. You