Generate Images and Videos Inside Claude (Higgsfield MCP Setup)

If you've been following along with the Claude series on the channel, you already know Claude can do a lot of things on its own. But until recently, one thing Claude couldn't do was generate images or videos directly inside chat. That just changed. In this video, I'm going to walk you through how to connect Claude to a tool called Higsfield, set it up in under a minute, and then show you

exactly what Claude can do once the connection is in place. No coding, no technical setup, just one connector you turn on. So, quick disclosure up front. This video is sponsored by Higsfield. They reached out to me and asked me to walk through their new Claude integration.

by Higsfield. They reached out to me and asked me to walk through their new Claude integration.

Higsfield is a creative AI platform. On its own, it's a website where you generate images, generate video clips, and edit media. That's been around for a while, and it's its own product.

What's new is that they built an MCP connector. MCP just stands for model context protocol. It's

how Claude connects to outside tools. You turn on the connector and Claude can now talk to that tool on your behalf. So when you turn on the Higsfield connector, Claude can generate images, generate videos, and edit media without ever leaving the chat. You ask Claude to make something.

Claude calls Higsfield in the background and the result lands right in your conversation.

One detail worth flagging up front is Higsfield gives Claude access to a few different image and video models. By default, Claude will pick one for you based on what you ask for, but you can

video models. By default, Claude will pick one for you based on what you ask for, but you can tell Claude exactly which model to use, and that's where things get interesting. Two of the strongest options are Cance 2.0 for video and GBT image 2 for still images. I'll show you both in action, and I'll show you how to specify them by name so you know how to do it yourself. Setup takes around

45 seconds. I'll walk you through it slowly so you can pause if you need to. Step one,

45 seconds. I'll walk you through it slowly so you can pause if you need to. Step one,

open up claw.ai in your browser. Click on the profile picture bottom left corner. Then click

settings. Step two, in the settings menu, find connectors on the left side. Click on it. So

this is the same panel that you'd see for other connectors. Step three, click add the custom connector. So a small window will pop up asking for two pieces of information. Step four, fill in

connector. So a small window will pop up asking for two pieces of information. Step four, fill in the two fields. For name, type Higsfield. For the URL, paste in the address https.mmcp.higsfield.ai.

AI/MCP. I'll have it in the description below so you don't have to retype it. Step five, click add.

That's the technical part. Done. The first time you actually use the connector, Claude will ask you to authorize it. That means you'll need a free Higsfield account and Claude will redirect you to sign in or sign up real quick. Once that's done, the connector stays connected and you don't have to do this again. You'll know it worked when you see the Higsfield in your connectors list and the

toggle next to it. It's on. Now, let's actually use it. I'm going to start with something simple just to confirm everything is wired up correctly. So, here's the prompt I'm going to type right here. Generate an image of a small bakery storefront on a quiet street in autumn. Warm

here. Generate an image of a small bakery storefront on a quiet street in autumn. Warm

light coming in from inside photographic style. Let's go and submit. Notice I haven't told Claude anything specific about which tool or which model to use. I just asked for an image. Let's see what happens. So, here's the result. It looks uh great. And then I want to show right here. This was by

happens. So, here's the result. It looks uh great. And then I want to show right here. This was by default created with nano banana which is a strong generalpurpose image model. And for most everyday requests you can just ask for an image and like claude pick the right tool. It's that simple. But

here's the thing. Higsfield gives you access to a few different models. And you can pick which one you want. Let me show you. So I'm going to give it this prompt. using the Higsfield MCP generate the

you want. Let me show you. So I'm going to give it this prompt. using the Higsfield MCP generate the same bakery image but this time use the GPT image too. So notice it's exploring the models. So two

things changed that prompt. I started with using the Higsfield MCP which tell Claude exactly what connectors to reach for and then I added use GPT image too so Higsfield knows what model to run.

And I can see right here it's generating using the GPT image 2. And here's the GPT image 2 version.

You can compare the two side by side and see the same prompt produces different looks depending on the model. Neither is better in some absolute sense. They just have different strengths. The

the model. Neither is better in some absolute sense. They just have different strengths. The

point is for you to have the choice and make the choice in plain English. No technical syntax, no settings menu. Quick recap of when you'd use each approach. If you just want a good image and you don't care about the model, ask plainly and let Claude route it. If you want a specific model because you found one fits your style better or because you want to compare results,

name it explicitly using the Hakesfield MCP prefix and then the model name. Now, let me show you one more thing that's useful. I can ask Claw to refine either of these images. I'm going to say this.

Take the GPT image 2 version, make it dusk instead of midday and add a person on the bench outside.

Now Claude sends another generation request and the conversation context carries through. I didn't

have to redescribe the whole scene from scratch and I didn't have to repeat using the Higsfield MCP or GBT image two parts Claude remembered. Now, that's the practical advantage of doing this inside Claude instead of switching to a separate tool. And there we go. We have Dusk and we have a person sitting on the bench. That's how easy it is to work with Higsfield within Claude. Okay,

let's try video. This is where I want to push the tool a little bit harder than just a quick clip. Seance can generate up to 15 seconds of footage, and it can handle multiple camera angles

clip. Seance can generate up to 15 seconds of footage, and it can handle multiple camera angles within a single clip if you describe them in the prompt. So, instead of a short 5-second loop, I'm going to ask it for a 10-second cinematic clip with three different camera angles that gives you a much better sense of what the model can actually do. A quick word on prompting video. The level

of detail you put into the prompt directly affects what comes out. With still images, you can get away with a sentence or two. With video, especially when you're asking for a camera movement, you really want to spell out things. you know, set the scene, describe the lighting, specify camera moves, call out all the things that you don't want. Uh, the longer the prompt

isn't busy work, it's the difference between getting something usable and getting something generic. Okay, here's going to be the prompt. It's about 1,200 characters. I believe when I

generic. Okay, here's going to be the prompt. It's about 1,200 characters. I believe when I was playing with sea dance before, I could get up to about 2,000 characters on it. So,

just to let you know what you can be work on, I'm just going to point out some of the key parts so you know what I asked for, you know. So, it's the same bakery scene we've been working with. Cobblestone street, autumn dusk, warm light with the windows, falling leaves for

working with. Cobblestone street, autumn dusk, warm light with the windows, falling leaves for uh throughout. Young woman on a bench reading a book. Uh then for the camera moves, I asked for

uh throughout. Young woman on a bench reading a book. Uh then for the camera moves, I asked for three distinct angles. A slow wide establishing shot from across the street. Uh push in over 4 seconds toward the storefront. Uh low angle, medium shot of the bench and a close-up detail of the leaves falling past the window. I also added the things that I didn't want. No text overlays,

no people walking past because the video models will sometimes throw those in and if you don't say otherwise. And um let's go ahead and go submit. Okay, so here's the finished result. I want to

otherwise. And um let's go ahead and go submit. Okay, so here's the finished result. I want to point out I didn't tell it to use seed dance. So just like before when the default was nanobanana, this went to clang 3.0. And I'm going to do it again because I want to generate this in seance. But we're going to watch this 10 seconds. Here we have some audio. You can see we have some

seance. But we're going to watch this 10 seconds. Here we have some audio. You can see we have some movement. We have a different camera angle here. And then we have our third shot. And then we have

movement. We have a different camera angle here. And then we have our third shot. And then we have our 10 seconds. Now let's try this again. Let's go to a new chat. And this time I'm going to do what I did before. I'm going to be specific using the Higsfield MCP generated 10-second cinematic clip.

Now everything else is going to be the same. I'm just telling it to put Cance 2.0 in it.

Now I can see it's picked the model that I want. We have seance 2.0. Quick honest heads up as this is running. Video generation is more demanding than still images and I've s found that seance can

is running. Video generation is more demanding than still images and I've s found that seance can occasionally fail on its first attempt. If that happens to you, just ask claw to try it again.

And in my testing, usually works on that retry. I want to show you the real experience here, not a polish version. So if it fails on this take, I'll just retry on the camera. But we

get success on the first go. Let's go ahead and hit play. [music] Definitely a music difference.

We have the person [music] sitting on the bench.

Feels a lot more cinematic than the first one. So again, just like looking at the pictures. I really

enjoy playing with C dance. I have a different video on that. I can link that. But h go ahead and try both and see what the quality you get and see what you like better. But remember the detail of the prompt. If something's off, the lighting feels wrong, or you wanted more leaves, the woman on the

the prompt. If something's off, the lighting feels wrong, or you wanted more leaves, the woman on the bench got cut from the shot, you can iterate. Tell Claude what to adjust, and it sends a new generation request. You stay in one conversation the whole time. The setup demos are nice, but the

generation request. You stay in one conversation the whole time. The setup demos are nice, but the real reason to do this is the workflow. So, let me show you something closer to how I'd actually use it in a day-to-day. Pretend I'm helping a small local coffee shop with a weekend social post. I'm gonna ask Claude to plan it and produce it all in one go. Here's the prompt. I run a small

post. I'm gonna ask Claude to plan it and produce it all in one go. Here's the prompt. I run a small coffee shop in the mountain town. Help me plan one Instagram post for this weekend. I want a still image and a short video clip. Keep the tone warm and inviting. Plan it first, then generate the assets. Once we agree on the plan, let's submit. Okay, so here's the concept that it's coming up

assets. Once we agree on the plan, let's submit. Okay, so here's the concept that it's coming up with. You know, slow Saturday at altitude uh still image. So, the hero ceramic mug of coffee short

with. You know, slow Saturday at altitude uh still image. So, the hero ceramic mug of coffee short video 6 to8 seconds caption Saturday's not in uh rush. They want a little bit more information. Uh,

do you want a specific drink? Uh, should I uh the background suggest inside the shop looking at the mountains? Any signature props? Mood preference. So, I'm going to give it some answers to all of

mountains? Any signature props? Mood preference. So, I'm going to give it some answers to all of these here. So, it's going to be the latte and I'm not going to read all this information, but it

these here. So, it's going to be the latte and I'm not going to read all this information, but it matches the questions it asks for more detail. And all I'm going to do is say use Higsfield at this point. Okay, it gave me a couple images to choose from here. And they're ask me to what to pick.

point. Okay, it gave me a couple images to choose from here. And they're ask me to what to pick.

I'm going to just say use the second one at this point before we move on to the clips. You can see it's using Nano Banano, but I didn't really tell it specifically what to use there either. Okay,

let's check this out. I can see we have Cance 2.0. Let's hit play. We have a little bit of sound in there. We have a 5-second clip. And this is the part that I think is genuinely useful useful for

there. We have a 5-second clip. And this is the part that I think is genuinely useful useful for small business owners. Anyone who's not running a full marketing department, you're not stitching three tools together. You're describing what you want, planning it with Claude, and Claude makes the media for you by connecting to Higsfield. Quick honest section. A few things to know

before we dive in. First, this is generative AI. The output won't always be perfect and you'll spend time iterating. That's normal. Don't expect first shot perfection, especially with video. As

I mentioned earlier, video generation can fail on the first attempt and working on the retry. That's

just where the technology is right now across the board. Second, Higsfield is a paid platform once you go past the free tier. Generating lots of videos in particular will use credits faster than image generation will. So, plan for that if you're testing seriously. And that's the walkthrough. If

you want to try this yourself, the link to create a free Higsfield account is in the description and pinned in the comments. You'll need an account before the connector will work and the setup takes a couple minutes. The connector setup steps in claude are exactly what I walk through in this video. Thanks for watching this week on Teachers Tech. I'll see you next time with more tech tips

video. Thanks for watching this week on Teachers Tech. I'll see you next time with more tech tips