Start a 1-Person AI Creative Agency with Higgsfield + Claude

What you're watching right now was made by one person in one day using two tools. We're talking a full brand identity, product photography, a dozen static ads, commercials, and a back-end system that runs the whole thing on autopilot. And the stack I'm about to walk you through used to take an entire creative team and 15 to 30 grand per month. But now you can do it all by yourself. With just two subscriptions, anyone can now start their own solo creative agency. People are now charging clients thousands of dollars a month to run their creative department remotely. And I'm sharing with you how you can set yours up today. Even if you don't want clients, but want to run creative for your own company, this video is going to cover it all. In case you're new here, my name is Liam Mley and I run the largest AI business community in the world with over 300,000 members. And at my agency, Morning.AI, we've built AI systems for Fortune 500 companies and NBA teams. And here's the promise to this video, guys. By the end of this video, you're going to have the full stack and the system wrapped around it, most importantly, to produce a full ad campaign for any client you take on. The stack is comprised of four layers. Firstly, Hicksfield, our partner on this video. So, shout out Higfield, which generates every image and every video for us. Secondly is Claude, which is going to be the brain for this operation. It's going to be writing the prompts, the briefs, and the copy, researches competitors, refineses the outputs, and orchestrates the overall pipeline. Thirdly, we have Notion, which is the agency backend where every client, every ad, and every approval is going to get tracked. And lastly, as an added bonus, we're going to be using Appify as a live competitive signal that feeds new ideas into the pipeline for our clients. So, here's the plan for this video. We're going to be starting off with generating the brand and product images for a fictional streetear client brand named Vault. Then we're going to build a second hero commercial for Vault using Claude with a prompt builder skill that I'm going to be giving you guys. Then we'll head into Higsfield's new marketing studio to use proven formats in literally minutes. And finally, we're going to wire all of this together into what I call an AIOS or AI operating system. We'll talk about the AIOS in more detail when we get there, but it's basically what's going to take you from generating a few and disconnected ideas to producing these full campaigns. And by the end of this video, you're going to have a complete workspace that runs the entire campaign for any client or for yourself and your business brand or products. All of the files needed to follow along with this tutorial are going to be linked in the description. You'll just need to join my free school community and head to the classroom section and grab the skill files, the AIOS workspace and the notion template. So, if you want to follow along, just pause the video now, go down and grab what you need and meet me back here and we can get started. Otherwise, let's get into it. So, before we start off, if you don't have an account yet, you'll need to sign up at higfield.ai. And once you log in there, there's going to be the screen and you're going to be greeted with the explore tab. We'll head over to the image tab to get started. And the community feed on the right is everyone else's work. So you can ignore that unless you want to scroll it for some inspiration. But the history tab on the top left is where your own outputs are going to land. Now let's say that our client Vault is launching a new product, a new boot, and they want to start marketing it. Obviously, since this client is imaginary, we don't have an actual image of this boot. So we're going to need to generate one to use for the marketing material that we're about to create. So this Vault brand, I see it kind of looking a bit like Balenciaga. So, while I could come up with a prompt myself to get this boot image and then feed it into Higsfield, that sounds like a job that's better suited for another key tool in our stack, which is Claude. So, if you don't have Claude yet, you're going to need to head to claw.ai and create an account. And I'll be showing you how to use the Claude desktop app for this. So, make sure that you download that and follow along. If you're already a Claude super user, you can run Claude in a code editor, and that's going to work too. That's my preferred way. But I want to make sure this video is beginner friendly as possible, which is why we'll be starting in the Claude desktop app. Now, over in the Claude app, I've asked it to give me a detailed prompt for the boot. I'm just going to roughly type in here what I have in mind for a futuristic kind of black boot with a hard exoskeleton shell. So the prompt is write me a detailed nanobana prompt for an all black concept of a sneaker boot hybrid construction soft technical inner hard exoskeleton outer editorial product photography on wet concrete no laces no logos. Then I'm going to copy the prompt and paste it over into the Higsfield image prompt input and I'm going to select Nana Banana 2 model which is one of the best right now. We want a square aspect ratio. Do a single generation and then generate. After a few iterations, this is the one that I locked in that I really liked for our boot. And so from this point on, this image is going to serve as our master reference for everything else that we make. And keep in mind, we had to do this step because we didn't have any images of our client product. Since this client is not real, they obviously can't send us any product images. But when you have actual clients, you'll typically be sent product images to work from, but you could even do the same step of generating some concept images for your client's new products or for your own products, too. Next up, we need the client's logo. And since they can't send that to us either, we're going to need to generate that, too. And I came up with the name Vault because that's the vibe our boot was giving us, you could say. And we're going to do a clean, simple text logo. And we're going to use the same method and get our prompt from Claude, which which I did earlier off camera. And I'm just going to paste it in here. I'm going to go with Nana Banana 2 again since it works really good with text. We're going to do a square aspect ratio. And then hit generate. Now, here's a quick tip. If you are iterating a lot, you can bump the batch size to four. And so four images per generation is generally going to be way faster than firing them off one at a time. And it will get you closer to what you're looking for a lot quicker. Again, after a couple iterations, I really like this one. So, I'm going to go with this. We need two more assets before we move on. To ensure consistency across the generations, we want a multiv- view reference sheet for the boot. And again, we typically ask for this from the client so we can view the product from all angles. So, I'll paste the prompt that I once again got from Claude and use the at tag to the boot reference image specifically so that the model knows exactly which asset I'm referencing inside of this prompt. This tagging method is especially useful when you have multiple images attached to a single prompt. By tagging the right image, as you mention, it means the model won't assume you meant another image that's also attached and accidentally waste your tokens on a wrong generation. So, this is essentially a variable for our reference image. Now, we'll hit generate. Honestly, this output looks pretty good to me. So, let's move on. Finally, the last asset we're going to need to mock up is of the boots packaging since later we're going to be doing an unboxing video, which is super sick. So, again, I got all of these prompts using the same method of prompting through Claude. And all of these prompts will be in the course resources below this video. So, I'm just going to drag the logo in, tag the logo, and since the boot is already attached here, the prompt now references both. We're going to do a batch of two and generate. And boom, that's our packaging all done. So now let's get on to the fun part and actually generate our videos. I'm about to show you something that until very recently would have required a massive budget to produce. So let's head over to the video tab. Now most people just dump in a reference image in here and just start firing off video generations and basically hoping for the best and that's a great way to burn through your credits. So before we generate anything, I'm going to go back to Claude to write the prompt for us properly. But before I do that, I want to show you guys a setup move that can actually save you a ton in your credits. So I'm dropping a skill into Claude. And in case you don't know, a Claude skill is basically like a small file that teaches Claude how to do one specific thing really, really well. In this case, how to write Higsfield prompts in the exact format that the platform wants. So you can just go to settings, capabilities, scroll down to customize, create a new skill, drop a file in, and we're done. Now Claude knows how to write Higsfield prompts without me having to do all the handholding for it. And again, this skill will be in the course resources linked below on my school community. Now to get my video prompt, I'm going to drop the assets into Claude. the boot, the logo, the multiv- view sheet, the packaging, and then I'll paste in the brief that I drafted earlier, which tells Claude that I want a Balenciaga Y2K kind of fashion film where the boot feels like a rare object from another world. So, the brief is for my commercial, I want a Y2K meets Balenciaga fashion film where the shoe feels like a rare object from another world with just a hint of fantasy in the background. No story, no dialogue, just striking visuals, sharp cuts in a cinematic mood across 30 seconds. I'm not going to bore you with brainstorming and back and forth, but after a couple of the concepts were rejected, we landed on the direction for the video commercial. Since Cedask 2.0 only generates 15 seconds at a time, but we want a commercial to be more like 30 seconds, I broke it into two prompts that will flow into each other. So, first I'll copy prompt one and paste it into Xfield. Attaching the reference images, of course. I'll select Seed Dance 2.0, a full version, not the fast one. 15 seconds long at 16x9 aspect ratio. We can drop the resolution down to 720p so we don't burn credits on this test pass. And then we're going to hit generate. Honestly, that's not that bad at all considering where we were like 6 months ago when you had to like generate these clips separately and then stitch them together in post. This already feels like a single cohesive scene just from one prompt. Now I'll grab part two's prompt and paste it into Hicksfield. attaching the boot again. But here's the key move. I'm also attaching the clip that we just generated as a video reference so that this new and second clip carries the visual language of the first one into the next 15 seconds. So I'll pull it in from the video generations here. Hit save. Now it's locked as a reference and then I can generate And that came out pretty well. I did iterate on this prompt a few times just to get there. And that's the reality of production. Uh you don't ship the first take. You're going to need to pick the cleanest batch and then move on. When working with real clients, they'll need a bunch of creative assets like UGC and testimonials, ads, product launches. So let's head to the marketing studio where Higfield gives us a range of presets. You can think of them as prompt templates trained around the kinds of ads that actually go viral on Meta and Tik Tok right now. So instead of wrestling with the prompt engineering yourself, you just pick a preset. You can feed in your product, assign it an avatar, and then you can generate. So, we'll go through three of these presets today. First, a UGC virtual try- on haul. So, quick definition, if you don't know, UGC stands for user generated content, which are those kind of casual influency style videos where someone tries the product on and then talks about it in kind of a a selfie style. A real UGC shoot can run you $200 to $500 per piece. So, we're saving the client quite a bit of time and money by generating it in seconds. Since it's for social, we're going to set the aspect ratio to vertical, which is a 9x6 with a 15-second duration. And I'll attach the boot as a product. For the avatar, Higsville gives you a list of presets or you can build a custom one from scratch or even just clone yourself. But I'm going to use this preset one here, Jaden, because he fits a demographic of our imaginary brand. So, we can click generate. All right. These vault boots look insane. I'm going to see how they fit. Okay. Wow. Super snug around the ankle. The sole is chunky but comfy. I'd wear these every day. All right. So, that was pretty good. and we didn't even feed it a prompt of what to say. So, UGC that would have cost $500 can now be generated for $5 in like 5 minutes. And these aren't random presets. They're built on prompt structures that have already gone viral on Meta. So, this is gold for marketers who want a stack of ad variations to test without wrestling with influencers or agencies. Now, let's say our client wants a product unboxing video. So, for that, we can use the marketing studios unboxing preset. We'll use the same product and the same avatar, but I'm going to be dropping in the packaging mockup that we generated earlier so that the model unboxes our actual packaging. Bro, the vaults are here. Look at this all black build. The cage is crazy. These feel so solid. I'm obsessed. Wow. Pretty solid. I did have to run a few iterations until I got the output I just showed you, but still a massive savings in time and effort. And lastly, we're going to be using the Hypermotion preset. And this is like a high-end product reveal preset. The kind of commercial that you see from the big brand ads. So, let's generate and see what we can create. So, everything we just did was for one client, and that's a fun side project, but the real play here is in stacking multiple clients at once. If you can get five clients that are paying you $2,000 a month, you're already at your 10K a month, which is a pretty common goal for people. But to run a solo company like this, it'll start falling apart without a system that manages the whole thing. And that's where an AIOS comes in. So an AIOS is a term that I came up with that stands for an AI operating system. And this is like a wrapper around your business, supercharges everything you do in it with AI. And it can be applied to any type of business, but it is particularly well suited to running a client-f facing agency business by taking the repeatable work off your plate. So you spend less time working in the business and have more time to work on the business with the bandwidth to actually scale. So let's set up our AIOS workspace. To make this beginner friendly, we're going to start by setting up our workspace on Claude co-work. After this point, we've been working in the chat tab, which is really just like chatting with any LLM like chatbt, but it couldn't take any actions for us. And the co-work tab is different since it can read and write files in your machine, and it can run tasks for us, which is exactly what an AIOS needs. So, I've created an AIOS folder for a creative agency, which you can download from the course resources. And while an AI operating system might sound complex, the core of it is really just a folder that lives on your computer with some useful stuff inside of it. You can think of it like a virtual desktop with tools you can use to perform tasks, a filing cabinet for organized storage, and a phone to call out for more info if you need. Once you unzip this AIOS folder I'm giving you, you'll see that it includes skill files, commands, a client folder, brand templates, and a setup guide that will help you to install it. I'll point to this workspace by creating a new project using that existing folder and name it whatever you want, and then hit create. Remember how I mentioned that the AIOS includes a setup guide? That means I can ask it to help me set this workspace up, and it's going to run a one-time interview. There's going to be five conversational questions about you, your agency, the formats you ship, your brand vibe, the aesthetics, and then it writes those answers into a file called aboutme.md. You can see the files here on the right. This is your visual reference for what files you're accessing as you're having a conversation with Claud and coowwork. So, as we answer our questions, these files will be updated in real time. During this setup, I'll tell it that I'm a solarreneur who helps small businesses with their content with AI. Now, I'm just speeding up this part for your convenience really, but the answers here would vary for everyone, of course. Just share what's relevant to you and what you hope to achieve. And after all of that, our workspace is all set up and ready to go. Before we run any client work through the AIOS, we need to do one more important step. We need to wire Higsfield into this workspace so Claude can fire generations for us directly. And that happens through the Higsfield MCP, short for model context protocol. An MCP is a way to plug an external tool into Claude so that Claude can call that tool like a function. So you can think of it as Claude getting hands. In this case, the hands have paint brushes and a video camera. And fortunately for us, Higsfield has their own MCP and we're going to use it from here in this tab. I'll copy in the URL from step two and then I'll go into my Claude settings. I'll go to connectors, customize, and then this plus button, add custom connectors here, and then we'll name our new connection. And we'll call it Higfield and paste in the URL from the Higfield tab here. Click add, then press connect and allow here. Now our Cord account is connected to Higfield. This long list of all the tools we have at our disposal now with this Higfield MCP. And remember, these are now the tools that we can use inside of our AIOS. Now, let's actually use this MCP. So, I'm going to drop in our boot here in co-work and tell it that I need one UGC and two high promotion videos. One's going to be like aquatic themed and the other is going to be like a fiery type of video. So, then send it's asking me a bunch of questions which is helpful. And I'm going to tell it to use the marketing studio presets like we did within the Higfield site earlier. And boom, just like that, we have our own little mini Higsfield running in our co-work. And the great thing about using Hicksfield in the co-work tab is how you can see your generations running and have some UI elements to do things like predict the virality or recreate it if it wasn't quite right the first time either. As a side note, it's also making the prompt itself based on all the information that has available within the workspace. Let's check the videos now. I literally stomped a puddle still felt locked in. Best all black boot I've tried. So, this is great and we're already ahead of the curve because we're able to generate multiple assets in one go right here from this chat, which has all of the context of our business and whatever we add into the client's folder about their brand, their customers, and so on. If you capped off your system here, it would still be highly useful. But there are a few limitations of using Claude plus Higsfield in this way because you have to download these assets yourself and manually attach the reference images in our chat. So, I'm going to show you an even better way of using Higsfield with your AIOS workspace, which is using Claude code. And I know for some of you, hearing the word code might feel intimidating, but it's as simple as just clicking the third tab here next to co-work. And after all, co-work is really just a simplified version of Claude code. And Claude code is just more powerful. and super users like myself and my friends and other business owners that I know are using right now. Commonly, people will use Claude Code inside of a code editor like VS Code. But using it from the desktop app here works well for most of our needs. Since we already gave Claude access to Higsfield by MTP, we can run Higsfield from within Claude Code 2. So if I drop the boot image and the logo into the chat and tell it this is my client Vault, a streetear brand. It can site getting to work. I'll ask it to generate 10 static ads in a 3x4 format using Nanabanana 2 with relevant text hooks and CTAs. You can see Claude reading the Higsfield skills as it works. And once it finishes, I can head into the AIOS folder. Once again, you can run your entire business from this workspace based on how you configure it. And that includes any context about who you are, what you do, who your clients are, organized neatly in their own folders with their own context and assets, your templates, and your plans as well. But right now, we're going to open up the clients folder. And as you can see, Claude already set up our vault client folder. So, if we go to projects and then meta ads, we'll see our renders that Claude saved directly to our computer for us. So to loop back to our desk analogy, it's like Claude took the video generator tool off our desk, used it to produce some videos, and then stored them in this filing cabinet inside of a folder for our own client all on its own. The boots right, the logo's right, the pallet's right. We didn't have to write a single individual prompt for these. Claude wrote them, find them at Higsfield via the MCP, and drop them here right in our folder for us. Now, of course, we could push this further. For instance, we could spec out a more detailed brand brief, install a dedicated copywriting skill. You can imagine how far we could take this by building this out for our own exact needs. But this is the power of your AIOS. And I'm telling you that this little Claude code chat window here is your gateway to insane levels of productivity. Running a creative agency yourself means managing a content calendar with all of your clients, syncing up with the right context about their brand, pulling in data on how each ad performed, and so on. So, let's take the last but very important step to sync up our AIOS to Notion. And that way, we're going to have a clean way to manage all of our clients content calendar and store everything in a database that tracks the performance of each asset we produce. So, our AIOS can iterate and improve over time. This notion template will also come with the course resources below this video on school. And put simply, it's just a visual database for all of our creative assets. This notion space is going to help you to track all of your clients. And as you can see, each client gets its own column. Every ad becomes a row. And every row, it attracts important info about that creative, such as the ads hook, copy, the visual prompt, and important KPIs like click-through rate, return on ad spend, and things like that. With this notion set up, we have a full backend to our AIOS that allows us to visualize our clients work on a calendar to keep ourselves organized in delivering work on deadline. AIOS can read all of this to remind us what's due. Analyze how ads perform so that it only repeats and refineses what it knows works and it becomes a self-improving organism that feeds off the past performance. We just need to connect notion to Claude. So we'll go to settings connectors browse connectors and then just search notion then add and continue. Now our workspace is connected to our notion. just copy the notion database link here and then tell Claude to push everything we have built so far into this notion ads database as a new client called vault and specify that this database URL that I just copied is where I want Claude to add those assets. Claude then reads the notion database schema to understand its structure and then adds every asset we generated into a new row. So if we go back to notion, there's now 10 new entries under our vault client column. Again, I'm doing all of this right from chatting within the AIOS workspace. And if if I open up one of these ads, you can see the image we generated from Higsfield MCP is also linked there. We're getting into pretty powerful territory now, but let's push it a little bit further. If we could monitor what our clients competitors are doing, we can use that knowledge to fuel our own workspace. So, let's look into scraping live competitor data for our clients. This functionality will allow our AIOS to quickly iterate on what's working for our competitors and generate hundreds of creatives right here from this chat in our client style. To achieve this, we'll hook up the entire workspace to Ampify, which is a super handy web scraping platform that lets us scrape things like Tik Tok and Instagram and YouTube and Meta Ads library, Reddit, basically anywhere you'd want to pull live signal from. And it's cheap, it's reliable, and crucially, it ships an MCP as well. So, same idea as before. Claude can drive it directly from within our workspace. There's no need to actually go on to Appy and click around. We can just tell Claude what we want and it fires the right scraper for us automatically. So, I'm going to go to console.appify.com. Sign up if you don't have an account, of course, and you'll get $5 of free credit as of this recording. I've already done that, so I'm going to go into the dashboard, click create, and then create a new token. Then just copy my user ID and token here and ask Claude Code to set it up for me. Now, let's take it for a spin. I'll tell it to use Appify to find Ba's competitors, scrape the competitor's trending, and high performing posts. Draft 10 ad copy ideas based off what's working, and push them all to the Notion database for review. Watch how Claude reads the data in our workspace and notion to pull Vault's brand context. The scraper kicks off. Ampify is now scouring competitor accounts and ad libraries, pulling back what's actually performing in this niche, trending angles, hook patterns, ad formats, source links on every claim. And when it's done, you can see it gave us 10 new ad drafts to review. Each one has a hook, the format type, the visual prompt, and a comments field saying which competitive pattern that it came from. This is all reflected over a notion. So, I'm going to tell it to fire these off. I'm also telling Claude where the images it needs are located on my computer so it can pull into my AOS. So, this is the one time setup. And then from here, my workspace will know which image I'm referencing without having to attach it on every chat. So, let's see the output from it. Works perfectly. And we have a UGC in our campaign with our avatar here. So, let's check that out, too. What is going on, guys? It's your boy Jason back with another video. And today, we're doing a quick review on the All Black Vault Tactical Boot. They're pretty sick. I like the design. I like the little strap on the side. So, there it is. You now have a full creative agency stack. brand assets, commercials, UGC, a competitor signal loop, and a notion backend that tracks and feeds it all. And the system is live and ready to work for you. But here's the thing, we only just scratched the surface of what this AIOS can do. So you could sync this up with tools like Slack or Linear to manage client communication and deliverable timelines without having to switch tabs. You could connect a tool like Make or Zapia to automate your reporting to the client. You could pipe Google Ads or Meta Ads performance back into Notion, so your AIOS is analyzing what actually converted for the client. Every one of those connections make the system smarter, more autonomous, and more valuable ultimately to your clients. And eventually, what you've built becomes self-improving where every scrape and ad and every result teaches the system what to do more of. And that's the real unlock here by using an AIOS. But here's where I want to be real with you guys for a second. Barrier to entry, as you've seen, for all of this is genuinely low. Like a computer, a few subscriptions, maybe a couple hundred a month, and a relatively creative mind. That's it. And that's exciting, but it also means that the stack itself is not your competitive advantage. Anyone can copy the stack now after this video goes out. And your edge is what happens before and after the generation, which is finding the right clients, closing them on retainer, negotiating contracts to protect you and your margins, setting expectations so clients don't come back to you after 2 weeks asking why their ads aren't printing money. And that is the real skill set, the business skill set. And without it, you can burn your reputation fast. One client who feels misled about results, one contract without proper terms, or one revision cycle with no boundaries, and suddenly your case study is like a cautionary tale to stop other people working with you. So, if you want the full picture and not just the tools, but the business that runs over top of them, head over to my free school community. The link is in the description along with all the resources from this video, the skills, the AIOS folder, the notion template, you'll find complete paths on how to land your first client under your new agency. And if you want to skip straight to the final line, there's also a link to check out my accelerator where my team and I will help you to fast track you to the level of success that you're actually after with the strategy, support, and accountability to get there. Because remember, like I said at the start of the video, all it takes is a few paying clients, 2, 3K a month each, and you're at that 10K per month level, which is really, really a huge achievement and the gateway to many, many, many more successes in online business. And the system to actually do the work is what we've just built here. Now, you just need to focus on getting the clients to actually run it for. So, if you want to learn more about how AI operating systems work, you can check out this video here. But aside from that, guys, that is all for the video. Thank you so much for watching and I'll see you in the next