Automatically generate any number of images and videos!? How anyone can link Claude and Higgsfield via MCP.

Hello! This is Sakista!

I'll say it right from the start.

You will be freed from the back-and-forth between chat AI and image/video generation AI.
You can execute and generate the latest image and video AI models directly from Claude!

You can build an environment where you can leave it to Claude to generate 10, 20, or as many patterns as you like.
Whether it's Nano Banana, GPT image 2, or videos...!

Prompt:
Analyze trending posts and generate 20 image patterns for Instagram using GPT image 2

I help out with fashion and beauty brands.
I usually work mainly on image and video generation. I also do vibe coding.

Because I use these in real-world settings, I am constantly searching for the best workflows and updating my methods daily.

In this note, I will talk about MCP, which will evolve your image and video generation AI workflow!

Actually, the other day, the image and video generation AI service "Higgsfield" released an MCP connector!!!

What is MCP?

MCP stands for Model Context Protocol, and simply put, it is a common rule for connecting external services and tools to AIs like Claude!

It's like a USB (common standard) specifically for AI! lol

In this case,

Claude ← MCP → Higgsfield

You can connect them in this way.

It's the image of being able to use Higgsfield's image and video generation from within a Claude chat. How wonderful...

Suppressing my excitement, let's proceed with the connection first.
I will explain it with screenshots, so please try it out yourself!

How to link Claude with Higgsfield using Higgsfield MCP

First, let's open the official Higgsfield MCP page.

The connection method is described step-by-step.
*There is an MCP link in the Higgsfield header.

Adding the Higgsfield MCP connector to Claude

We will proceed with how to add the Higgsfield MCP connector.

First, please open Claude.

From the Claude menu, go to Settings > Connectors.

On the connector screen,
select "Add custom connector".

Here is the information to enter into the form.

Name: Higgsfield
URL: https://mcp.higgsfield.ai/mcp

Finally, press Add.

Higgsfield is now displayed in the connectors. It's that easy!

Next, please press Connect.

Regarding permissions

The details are as follows.
Please read the content carefully, and if there are no issues, press Allow.

Permissions granted to Claude
- Identity verification
- Access to basic profile information
- Access to email address

Warning:
Please ensure that you trust Claude (claude.ai).
You may be sharing sensitive data with this site or app.

It has been added successfully.

Configuring each tool

Optional settings

The Allow, Always allow, Block settings for each tool
should be configured based on your own workflow.

You can also configure permission settings in the middle of a Claude conversation.

How to use Higgsfield MCP / Claude

Generate images

By connecting Higgsfield to Claude, you can generate images directly from your Claude conversations!

Creating images right from within a conversation... Instead of the usual flow of brainstorming, writing prompts, and moving to a separate tool to generate, the key feature here is being able to proceed all the way to image generation while talking to Claude.

Wait, couldn't you just use ChatGPT with GPT Image 2 for that?

I'm sure many of you felt that way!

There are many benefits to using the Higgsfield MCP.

5 benefits of generating with Claude using the Higgsfield MCP connector

1. Seamlessly use various image and video generation services that have API integration with Higgsfield

2. Leverage Claude's capabilities, memory, and skills to generate content

3. Start generation in the middle of a conversation while maintaining the context of your discussion and ideas

4. Make changes or corrections to generation results directly in a conversational format. You can even generate videos from there

5. Capable of autonomous generation following Claude's instructions

Number 5 is especially convenient. It allows for generation workflows that are difficult with Higgsfield alone.

For example...

Analyze thumbnails that rank high for XX, then generate 5 patterns of thumbnails that are likely to be effective
Generate 20 patterns of UGC-style images using the product -> then generate a video
Use Higgsfield to generate the images needed for the site I'm building in Claude
Think of storyboard ideas needed for this story, then generate images with Higgsfield

The ideas are endless, aren't they?

The introduction has been long, but
let's start by simply generating an image using the Higgsfield MCP.

Prompt sent to Claude

When generating images during a casual conversation like this,
Claude automatically sends an image generation prompt to Higgsfield.

Generation has started within Claude.

Huh? This isn't the image I was expecting...

Looking at the Claude conversation history

Since no model was specified, the Higgsfield model: Soul was selected.
An image with a different nuance than the instructions has been generated.

I will continue the conversation.

Generate using Nano Banana

I am giving instructions in the flow of the current conversation.

Here is the generation result

Wow🤍 It's close to the image I roughly imagined!

Wait, which one is it? lol
It's probably Nano Banana 2.

Unlike Soul, it is faithful to the prompt. As expected of Nano Banana.

Next, I will try it with GPT image 2.

Here is the generation result

GPT image 2 is also lovely...🤍

About credits

Credit consumption is as follows.

・Unlimited models (varies by plan): No credit consumption
・Paid models: Standard credit consumption

Note: Resolution is generated at 2K by default if not specified. If you are concerned about consumption, please select 1K. *2K free models do not consume credits.

Also, the quality is set to the highest level by default.
For example, in the case of GPT image 2, "High" is selected. If you have a preference, be sure to specify it, such as "Medium".

Claude sessions, naturally, consume credits based on the conversation.

Making changes to images

Next, I would like to try making changes to the generated image.
I tried instructing it to change the hair color.

Instruction to Claude

Here is Claude's response:

Image generation result

Claude's response:

You can also click on the image to continue the conversation

Clicking on the generated image opens a dedicated chat screen.

Prompt

Here is the generation result.

How to find advertising angles and generate videos and images

Next, I will try generating creative assets for advertisements.
It is very convenient because you can brainstorm and refine ideas while bouncing them off Claude.

Prompt

Claude's response

Here is Claude's response.

Generated images (conceived by Claude, generated by Nanno Banana)

Since this is Text to Image, this alone is naturally not enough.
This time, I will use it as a rough draft.

I will reflect the brand's model and product onto this rough draft.

Character Cheat Sheet

Product Kit

Please refer to the following note for how to create the sheet.

I have referenced the model and brand kit to the rough draft generated earlier.
I will explain the referencing method later.

Image generation results with the product image reflected

By the way, I generated a thumbnail for EC.

Thinking of a story with Claude and generating images using Higgsfield MCP

Next, I will explain how to have Claude think of a story for a video based on a single image, and generate multiple images along that story.

By expanding the worldview, characters, and scene flow from a single image, you can collectively create materials that are easy to use for videos.

First, prepare the base image.

Have Claude generate ideas.

Prompt

I had it create 8 ideas.

I will send the next prompt.
It is very convenient because it sends image generation instructions to Higgsfield while understanding the context.

How to reference images

When referencing images, please instruct it to use the Job ID.
Example: Please be sure to reference e8xxxxxx.

Whether the Job ID works for Claude, or please reference the image URL.
The image URL contains the date, number, and Job ID in that order.

By the way, it is also possible to give instructions like 'Please generate a new image using the image generated earlier' based on the flow of the conversation.

Images automatically generated in 8 patterns using Higgsfield MCP from Claude

The image generation model is GPT image 2.

Video generation

Using the generated scene, I created a video with Seedance 2.0🤍
This vlog-style video was easily generated by using Higgsfield MCP.
It feels like I can create any work with Claude as long as I have a small idea.
It is quite shocking because I can use the full power of Claude.

See you in the next note🤍