Transcription
All right, so Higgsfield just released their MCP, which now means you can use Claude Code to generate AI images and AI video without leaving the Claude app. So, in this video, I'm going to show you how to connect Higgsfield to your Claude account, and we're going to go through a workflow where we're going to use Claude to generate assets for an e-commerce campaign like this, this, this, this, >> [music] >> this.
"This hoodie feels like a weighted hug, and it's the one I keep reaching for." >> and this. "Just feel the weight of this. It grounds me. That's why I never take it off." >>
Not only that, I'm going to show you how you can use Firecrawl to go out, scrape your website, and come back with a brand bible that Claude can then reference in every single prompt that it feeds to Higgsfield. Speaking of the prompts, if you want access to all of the prompts along with a bunch of other AI workflows and tools, check out the link below this video.
All right, to set this up inside of Claude Code is pretty easy. We need to open up our settings, add the custom connector, and then authenticate by logging into Higgsfield. So, let's first open up our Claude settings and go to connector. So, I'm going to go back into Claude Code. I'm going to click this little button up here, and then I'm going to click on my name. I'm going to go to settings. Then I'm going to go to connectors. Then at the very bottom, I'm going to click on add custom connector. Now, we want to give this a name, so we'll do Higgsfield, and then we need to paste in that server URL from the Higgsfield page where it says add a custom connector. So, I'm going to copy this URL. I'm going to go back to Claude. I'm going to paste that in, and then I'm just going to click on this add button. Now, what it's going to do is it will add the MCP to our settings, but we're not yet connected. So, what you want to do next is click this connect button. It's going to open up a new tab right now. So, it's basically authenticating your Higgsfield account to Claude. Make sure you're signed into your Higgsfield account. I'm going to click on allow, and then that's going to refresh my Claude. And now when I go down to my connectors, we can see Higgsfield here is connected. So, that's how you get it set up.
All right, now, before we start generating images and assets in Higgsfield, we want to use this with Claude. So, what I'm going to do, I'm going to use Comfort as our use case in this video. They make the hoodies. So, what I'm going to do is again, just to demonstrate the multi-step capability of having Claude and Higgsfield together. I'm going to have Claude do a little bit work. I'm going to say, "Hey, can you research comfort.com, pull the brand voice, hero products, visual style, and target customer, and then build a brand brief that I can reference for the rest of this campaign." I'm going to send that off. I'm going to trust this workspace, and I'm working in a specific folder on my desktop so that Claude can generate the output in that folder. So, you can see this is the folder that I created on my computer. Right now, it's empty, but whenever Claude generates output, it's going to go into this folder. So, right now, what it's doing is it's using Firecrawl, which can be a separate video. I have the Firecrawl MCP set up as well, which is going to go out and scrape the Comfort website and pull in all of the branding information, target customer, all of the products. While that's working, let me just call out Firecrawl real quick. So, most of you are probably familiar with Firecrawl. Again, topic for another video, but it's basically a web scraping tool that also has an MCP. So, we can give Claude access to this tool, which will go out and scrape any website that we want and pull in a bunch of information that we can prompt in natural language. So, this is the tool I'm using to scrape the Comfort website.
All right, so we can see Firecrawl has done its work, and now Claude is using the output to generate this brand brief for Comfort, including a brand snapshot. We've got a brand voice. It's building a nice table here for their hero products, pricing posture, visual style. So, Firecrawl, if you have not used it, is definitely cool. They also offer some free credits if you want to get started for free. I'm not sponsored by them or anything, but I do use them in Claude Code quite a bit. But, we can see here Claude is still going. We've got our target customer. We've got our differentiators, campaign-ready hooks. And you can see in the output, we all we you know, we have the visual style, which includes the logo. We've got the color palette here. So, all of this is going to be really good information for the assets that we generate in Higgsfield.
Now, for the next step, let's see if we can use Claude with Higgsfield to generate a static meta ad using the new ChatGPT Images 2.0 model. So, what I've done is I've taken a reference image from the Comfort website. So, this image of this woman wearing this sweatshirt, and I placed it in my project folder here. So, you can see I have a folder called source assets, and then under there I have images, and I've placed an image of that hoodie directly in there so Claude can access that. So, what I'm going to do is go back to Claude, and I'm going to paste in this prompt. And I'm going to say, "Generate a hero static ad for the Comfort Signature Hoodie using GPT Image 2.0 with the Higgsfield MCP." And then I'm pointing the reference image, pointing Claude to the reference image in the project folder. And I'm going to say, "Hey, pull the brand brief for visual style, mood, and creative direction." This And I'm saying, "Hey, match Comfort's clean lifestyle aesthetic, feature the hoodie as the clear hero, use 9 by 16, don't do any on-screen text." You can kind of prompt this however you want. And I'm saying, "Hey, generate four variations so we can pick the strongest. After generation, evaluate each one against the brand brief and recommend the strongest variation with reasoning." So, again, it's not just about using Claude to use Higgsfield. It's using Claude as part of the creative process where it's not only going to send the prompts, but it's going to actually review the output for us against the brand guidelines. So, let's test sending this off and see how Claude handles the MCP request to Higgsfield. So, it submitted the request for all of the generations. Now, it's just waiting to get the results back.
All right, so you can see that Claude has finished up generating those images. One of them got stalled, and it reran it on its own. Now, it's downloading them to the project to review them to see which one might work best in a static ad. I'll show them to you in a second. All right, so we just finished up, and there's a there's a lot going on here, which is really cool. So, first let me show you the downloaded output. Let me go back into our folder, and now we have this output folder, and then we have the Comfort Hero variations. And here are the four images that Higgsfield generated, right? So, we took the image of the model, and it basically made variations of her in different settings. So, you could not only use this in a static ad, you could use this on your Shopify website showing off the model. So, that's her in a bedroom. That's her in an armchair. I kind of like this one, actually. Let's look at the outdoor one. So, this is her outdoors. It kind of looks photoshopped on there. I'm not going to lie, but that's the third one. And then let's see what they came up with for studio editorial. All right, I actually think this one looks pretty good, as well. So, that's the four outputs.
Now, what we asked Claude to do was evaluate each image against the brand brief, okay? So, it's saying the variation evaluation, each scored against the brand brief, and it took each image that Higgsfield generated, and it came up with this scoring system. And it's recommending version three, which is the outdoor one. Actually, the one that I like the least, but it's saying for a paid hero static destined for reels or stories, V3 is the strongest pick. Here's why it wins, okay? So, we're able to have Claude run the generations in Higgsfield for us, and then actually review the output.
Now, real quick, cuz I know a lot of you are wondering about costs. As I mentioned, I have a paid Higgsfield plan. I'm on the creator plan. I'm not sponsored by them. I actually pay for this tool. And when we go into the credits, right? So, Higgsfield runs on a subscription base, and with each subscription, you get a certain amount of credits. Those generations through the MCP, each image costs four credits each. And not only that, inside your Higgsfield account, what you can do is you can go to this assets tab, and you can see all the assets that you've generated in Higgsfield. And here are those four images. So, you can see them in your Higgsfield account. You can see the prompts that was used. You can see the quality. We did not specify. You can obviously specify that. We have highest quality here. So, if you ever want to check the output in your Higgsfield account, just go to the assets tab.
Now, this output is pretty good really as a hero shot. If we wanted to try to turn this into a static ad with text, we could certainly try and have Claude orchestrate this with Higgsfield. So, I've got a prompt here that's saying, "Hey, can you take V3, the one that you recommend, use GPT Image 2.0 to add copy overlay." So, I'm saying same image just with text layered in, and then I've got a bit of a prompt here about what we're trying to add in terms of the text. Again, just seeing if Claude can handle this as the orchestration layer. So, let's fire this off and see if it will take our new generation, which is her outdoors, and add some text on top of the image.
All right, so Claude just finished up. It says it created the variants. I haven't even opened them up yet, but it's giving us kind of a summary of each image, and it's giving us their recommendation, which is variant B. So, let's go back into our project folder and you can see here we've got these two new files here, these version three with the overlay. So, let's look at that one. All right, the font is definitely way too small on that. Let's look at the second one here. All right, a little bit bigger. The bottom part looks good. The top clearly is not great. We could definitely iterate back and forth if we wanted to. So, again, we could go back and forth to obviously make this font bolder and bigger. I do think the rest of the image looks pretty good, but let's keep going.
Let's now go to the next step, which is we're going to animate a hero shot with C Dance 2.0. So, we want to take the V3 from the original output and use C Dance just to see if we can animate this. So, what I'm going to do again is paste in a prompt. So, I'm going to say now animate the hero static using C Dance 2.0 using Higgs Field. I'm going to point it to the reference image. Generate a 5-second cinematic motion shot. The opening hero clip of a premium DTC real ad. And then Claude gave me some direction here. Slow, subtle camera pushing, gentle pastel sky drift in the background. Hoodie fabric catches natural light as the camera moves. Model holds the pose with the minimal motion. All right, so we've got some direction here from Claude and I'm asking it to save the output to the folder. And then once it's complete, evaluate the clip against the brand brief. So, let's fire off that prompt. Let's see if Claude can use Higgs Field to animate this image with C Dance 2.0.
All right, so that finished up and I noticed it faster inside of my Higgs Field account than inside of Claude. So, I just went into my all assets and noticed the video. So, let's take a look at that. >> [music] [music] >> All right, pretty good animation there. C Dance added some music. You can see we're using the C Dance 2 model 1080p and C Dance FYI is not a cheap model. We can see that cost 45 credits. All right, and now Claude is finally catching up. I guess Higgs Field took a second to get the information back. So, you can see it's downloading the file. It's saved it to our project. We can see hero animated and we can see the actual raw video file right here. Now, if you're wondering how is Claude actually watching and analyzing this video? I have the Gemini Vision API hooked up to my Claude account. That's a topic for another video, but I basically given Claude the ability to watch and analyze video content. So, you can see Claude is doing its evaluation of the video here in terms of the fidelity to the still, meaning does the still image actually hold the fidelity through the video? And it's grading it against the brand brief checklist.
All right, so we've got our kind of hero image. We've got a static ad that needs to change the headline and we've animated our hero image with C Dance 2.0. Let's pivot a little bit and let's try to come up with a UGC angle here. So, first and again, this is all happening in the same conversation with Claude. No reason to go back and forth inside of Higgs Field. Let's have Claude help us generate a UGC creator who's wearing the Comfort sweatshirt using GPT Image 2.0. So, we're going to paste in this prompt here and again, we're going to ask Claude to pull the brand brief to come up with the primary persona. We're saying it should feel authentically native to TikTok and Reels. We've got a little creative direction here in terms of the prompts and we're asking for two variations and again, for Claude to review the output from Higgs Field to pick the strongest one. So, again, let's send that off and see what Claude and Higgs Field come back with.
All right, so Claude came back here. It created these two images. One is a brunette on a couch. One we've got a blonde woman in the kitchen. So, again, if we go into our folder here now we have a new folder called UGC creators and we can open these up. So, here is the first one. That is a really really good output from Chat GPT. Let's look at the second one and another very strong output. I mean, that looks amazingly real. I think both of those came out really well. So, Claude as we asked it to do, analyzed and kind of picked the winner here. So, it's going with creator B. It's saying the persona match is tighter. The emotional context sets up the hoodie reveal. And slightly tired, but content expression are the perfect before state for then I throw on this hoodie and reset my whole day. So, I do love Claude as the orchestration layer. Who's not only sending off all of the prompts, but it's reviewing the output against the brand guidelines.
All right, so Claude's going to go with creator B. I think I kind of agree there. And again, going back into my Higgs Field account into my assets you can see the output here with the full prompt that he used. So, let's go ahead and use this creator.
All right, we've got our creator. Now, let's try to take the image of our creator and put her into two different scenes with the hoodie using C Dance 2.0. So, I'm going to paste in a longer prompt here in the Claude. All right, so I'm saying, "Hey, lock in creator B. That's who we want to use. Now, generate two UGC video clips using C Dance 2 with Higgs Field." I'm pointing it at the reference images, which are the creator and also the hoodie. And I'm saying to use the brand brief. The voice should be warm, soft-spoken, self-care coded. And then we're going to try to generate two clips here. A native testimonial where she's sitting on her couch and wearing the hoodie. And then clip two, more of a product showcase where let's see if C Dance can handle this, where she's pulling the hoodie over her current outfit. So, two clips using C Dance, Claude as the orchestrator. Let's send this off and see how we do.
All right, so that finished up and we've got both clips saved into our project folder. So, you can see we have UGC clips here. Let's look at these clips. So, here is the first one. There is our UGC creator. There is our hoodie. Let's turn up the volume. "This hoodie feels like a weighted hug and it's the one I keep reaching for." All right, pretty good as a hug. We'd probably speed that up just a little bit, but in terms of the quality of the video, the realism the voice, pretty good output right there. We have clip two right here. So, there she is. She's kind of mid trying to put the sweatshirt on. Let's play this. "Just feel the weight of this. It grounds me. That's why I never take it off." All right, I think uh I knew we were going to have an issue with her trying to put it on. Let's play it again. "Just feel the weight of this. It grounds me. That's why I never take it off." All right, so you can see the the sweatshirt kind of you know, appeared on her without really going over her. Clearly an artifact there, but otherwise in terms of the quality of the character and the consistency of of the character. So, let me just show these side by side. You can see how good a job C Dance does keeping the creator looking exactly the same. Even the background looks exactly the same as well. And then again, we asked Claude to evaluate the clips. It's able to evaluate everything here except for the voice quality. It's not able to listen into the audio. Overall, hopefully you get the idea of using Claude as the orchestrator here. Kind of as a co-pilot when you're creating your assets. So, that's how you set up the Higgs Field MCP. That's how you use it with Claude. So, if you think back to what we built, we took a single product URL from the Comfort website. We used Fire Crawl to go out, scrape all of the brand information and just using a single chat and going back and forth with Claude we were able to generate some static images, animate those images generate UGC characters and then animate those characters with C Dance. So, overall, I think you can build some really good workflows with Claude and Higgs Field.