I connected one plugin to ChatGPT Work, typed a prompt, and had on-brand course mock-ups back for about 9 cents each. That plugin is the Higgsfield MCP. It lets ChatGPT Work generate images and video with a whole range of leading models, OpenAI’s among them, and you never leave the chat. Here’s how to set it up, which models to reach for, and what it actually costs. (Higgsfield sponsored the video this guide is built on. Every figure below is what I actually paid.)

What the Higgsfield MCP is

ChatGPT Work is a capable writer and researcher, but on its own it makes images with OpenAI’s model only, and it can’t make video at all. That’s the gap Higgsfield fills. It’s an AI media generation platform with a long list of image and video models behind it, and it makes them available through an MCP, the standard that lets an AI agent call an outside tool. Once you connect it, ChatGPT Work plays creative director: It writes the prompt, picks the model and hands the actual generation to Higgsfield.

What you gain is choice. Higgsfield offers image models like Soul and Nano Banana and video models like Seedance and Kling, each suited to different jobs. So you describe the result you want and pick whichever model handles that kind of job best, whether that’s a clean product shot or realistic motion.

Connect it to ChatGPT Work

Setup takes a couple of minutes, because Higgsfield is one of ChatGPT Work’s built-in plugins and there’s no URL to paste.

You’ll need a Higgsfield account first, on any plan, which you can create at higgsfield.ai. Then, in the ChatGPT desktop app, open Plugins on the left, search for Higgsfield and click the plus to add it. It asks you to connect your account, and if you’re already signed in to Higgsfield in your browser, that’s a single click. You’ll then see a confirmation that the plugin is installed and ready.

If you run on Claude, it’s nearly as quick: Go to Settings, then Connectors, add a custom connector and paste the Higgsfield MCP URL, https://mcp.higgsfield.ai/mcp. Higgsfield’s own guide to connecting an AI agent covers both routes, and an independent walk-through shows the same steps with screenshots.

Your first images

With the plugin connected, you call it from a prompt. I type the @ symbol, choose the Higgsfield plugin and describe what I want. For our Black Friday campaign, I attached a reference image and a couple of screenshots of the course and named the image model I wanted. Then I asked for 10 to 15 variations of a course mock-up on a transparent background. I left the ad copy off for now, since that comes later.

It worked for under 10 minutes and then showed its results in a small Higgsfield panel inside the chat. There were several directions: Some leaned on the course certificate, some made the course feel like a physical product, and some put it on a laptop screen. For each image you can see the exact references it used and the prompt ChatGPT Work wrote for it. That’s more useful than it sounds, because when one direction works, you can see why and ask for more like it.

Before you scale up, test. Higgsfield recommends a starting image model, and you can switch to another if the look isn’t right. So generate a single variation, check it, and only then let it make the other 200.

From images to video

You start with images because they’re cheap and quick to test, and you move to video once you know which message works. The same plugin handles both: You give ChatGPT Work a clip or a script, it sends the job to Higgsfield, and a video model like Seedance 2.5 does the generation. Turning a rough phone recording into a finished Meta ad takes a few more steps, so I’ve covered the full before-and-after in the guide to AI video ads.

The habit from the images carries over, and it matters more here because the video models vary a lot in price. Test a five-second clip on the model you’re considering before you commit to the full render.

Export to After Effects

If you have an editor on the team, this feature is worth knowing about. Higgsfield has a direct connector to Adobe After Effects, and using it costs no credits. You can ask it to open your generated video as an After Effects project, with the motion-design animations intact and fully editable. I’ve no idea how After Effects works myself, but that’s the point of the hand-off: You, or your editor, can fine-tune everything from there without rebuilding a thing.

What it costs

Higgsfield runs on credits. You pick a plan, it gives you a monthly credit allowance, and each generation uses some of it up. Plans start at around $15 a month, and the Ultra plan I used costs $129 a month, with larger credit bundles available on top. If you want every tier, Krea’s breakdown of Higgsfield’s pricing is the clearest I’ve found.

What each generation costs tells you more than the plan price does, so here are the real numbers from the video:

• Images: About 2.75 credits each on the model I used, which worked out at roughly 9 cents an image on my plan. Those 10 course mock-ups cost under a dollar.

• Video: On Seedance 2.5, about $3.23 per five seconds. My 40-second edited ad came to $25.80, and three seconds of B-roll added $2.08.

• After Effects: No credits at all.

Those are the figures at the time of filming, so check the current rates before you budget. The pattern will hold, though: Testing with images costs pennies and video is where the money goes, which is exactly why you test with images first.

Where to start

Connect the plugin, give it one real job, and look at the first result before you scale it up. Everything I built around it, the skills and prompts for research, headlines, images and video, is free in the 9x community, where our operators share the workflows they actually run. That’s a quick way to find out whether an AI ad creative system earns its place in your own work.

Common questions

What is the Higgsfield MCP? The Higgsfield MCP is a connector that lets an AI agent like ChatGPT Work or Claude generate images and video through Higgsfield’s models, straight from the chat. The agent writes the prompt and picks the model, and Higgsfield does the generation.

Is Higgsfield built into ChatGPT? Yes. Higgsfield is one of ChatGPT Work’s built-in plugins, so you add it from the Plugins menu and sign in, with no URL to paste. On Claude, you add it as a custom connector using the Higgsfield MCP URL.

How much does Higgsfield cost? Higgsfield runs on monthly credit plans, from around $15 a month up to $129 for the Ultra plan, with larger credit bundles available on top. In practice, images cost a few cents each and video costs a few dollars per short clip, but check the current per-model rates before you budget.

Which Higgsfield model should I use? Start with a recommended image model such as Soul or Nano Banana, and a video model such as Seedance for realistic motion. Higgsfield’s model guide suggests which to reach for by job. Whichever you pick, test one generation on it before you run a batch.