Claude Can Now Make AI Videos By Itself (Higgsfield MCP)

Author: Prompt Engineer 48 · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Claude Can Now Make AI Videos By Itself (Higgsfield MCP) · Channel: Prompt Engineer 48 · 08:48 · Watch on YouTube

Topics: AI agents, Claude, Multimodal models

Summary

The video demonstrates how to connect Higgsfield to Claude through an MCP connector and use chat prompts to generate videos without opening an editing app. It shows workflows for researching property listings, iterating on video generations, and keeping a consistent visual style across a series. The presenter also tours Higgsfield’s Cinema Studio, Seedance, Canvas, and Supercomputer features. Finally, the video reviews annual plan prices and per-generation credit costs, and notes that unlimited models and free generations are not available through MCP.

Key points

Higgsfield’s MCP connector can be added to Claude and authorized without writing code; the presenter says setup took about a minute.
Claude can use Higgsfield tools, including checking account balance and generating videos from chat prompts.
A property-listing workflow has Claude research homes, write camera directions, and generate videos; the presenter notes that the first run may require prompt corrections.
Video ideas can be regenerated with small prompt changes, and a shared style formula can help keep a multi-clip series consistent.
Cinema Studio 4.0 supports camera-move prompts, motion control, and a personal assistant that can help choose a camera, light a scene, write prompts, or break scripts into shots.
The Seedance 2.0 workflow uses an image reference, a natural-language prompt that can describe sounds, and a generate step.
Canvas is presented as a shared, node-based workspace connecting prompt, image, and video nodes; Supercomputer lets users describe a job and select from agent skills.
The pricing page lists Basic at $9 per month, Pro at $29, and Max at $59, billed annually; Max includes 3,600 credits per month.
The video gives example generation costs, including about six credits for a five-second Kling 3.0 clip at 720p and about 22 credits for a five-second Seedance 2.0 clip at 720p.
Unlimited models and free generations are limited to Higgsfield.ai and do not work through MCP, CLI, Canvas, or Supercomputer.

Chapters

Start	Chapter	Summary
00:00	Higgsfield overview and Claude connection	The presenter introduces Higgsfield’s AI tools and demonstrates connecting its MCP server to Claude.
01:33	Claude video-generation workflows	Claude checks the Higgsfield account and generates listing videos, while the presenter also shows iterative generations and a consistent style across a clip series.
04:10	Cinema Studio and Seedance	The video tours camera controls, product-ad examples, motion control, the Mr. Higgs assistant, and Seedance’s image-and-prompt workflow.
06:14	Canvas, Supercomputer, and agent skills	The presenter describes Canvas’s shared node-based workflow and Supercomputer’s agent skills and explainer styles.
07:16	Pricing, credits, and MCP limitation	The video reviews plan and generation costs, then explains that unlimited models and free generations do not apply through MCP and several other tools.

Quotes

51 interactive tools and 33 app only tools — 01:33

I did not open a single editing app. — 03:07

You decide the style one time, and Claude keeps it consistent across the whole series. — 04:10

So, the sound is part of the prompt. — 05:42

So, the chat workflow uses credits. — 08:18

Concepts and entities

Higgsfield, Claude, Higgsfield MCP, Kling 3.0, Cinema Studio 4.0, Seedance 2.0, Canvas, Supercomputer, Mr. Higgs

Full transcript (original language)

00:00 This is what one prompt in Claude can turn into. And this is what Claude did to get there. It called a tool named generate video on Higgsfield with Kling 3.09 by 16 10 seconds audio on. All right, in this video, I’m going to show you the whole thing on screen, how I connected Higgsfield to Claude in about a minute, what it made, what it costs in real credits, and the one thing it cannot do. First, what is Higgsfield? If you open higsfield.ai, this is the home page, and you can see the top bar already says a

00:31 lot. Image, video, audio, MCP, then cinema studio, marketing studio, canvas, and supercomputer. So, it is one place for AI video and AI image generation, and the models sit side by side. Right now, the big card on the home page says Sirans 2.5 Mark top. And next to it, you have Kling 3.0, Wen 3.0, and cinema studio 4.0. Now, there are two cards here with a green free tag, AI Influencer and Genjutsu. That matters because if you’re looking for a free AI video generator to

01:03 start with, these are the two places on this page where you can try it without spending credits. And this one here, Higgsfield MCP for Claude, that is the part I want to spend most of this video on because it changes how you work. So, how do you connect it? I recorded this on my own screen, so let’s go through it step by step. I opened the Higgsfield MCP page right here. You can see the production skills bundle for Claude and the orange button that says try in Claude. I click it and Claude opens an add custom connector window. The name Higgsfield and the address

01:33 mcp.higsfield.ai/mcp are already filled in, and authentication is set to sign in now. Then Higgsfield asks me to authorize access. I approve it, and back in Claude, you can see the message at the bottom right that says connected to Higgsfield. And now the tool list opens 51 interactive tools and 33 app only tools, add studio brand, add studio creatives, create video analysis, check job status, and many more, each with its own permission switch. To test it, I just typed is Hicksfield connected. Claude

02:05 used the check balance tool, and it answers that Hicksfield is connected and working on the Ultra plan with 24,300 credits available. That is about a minute and no code. Now, let’s see what this does in a real workflow. It starts with one sentence in the chat, find the most expensive homes in a city, collect their photos, and generate a video for each one. I’m only showing you the Claude side on the left. Claude reads the listing, writes a camera direction, and then calls Hicksfield generate video. If you look here, the prompt says, “Slow cinematic crane up over the

02:37 canyon villa terrace at golden hour.” And below that, you can see the settings it picked itself, playing 3.09 by 16 10 seconds with audio. Then, the box just says generating. While the videos are rendering, Claude keeps working. It says, “Before pricing, one search to close a gap.” So, it is still researching the next step while the render runs in the background. And these are the results, the villa at sunset with a pool and the mountains behind it, first as a still and then in the player as a finished vertical video

03:07 you can download. These came out of one conversation, and I did not open a single editing app. Now, one honest thing this listing workflow needs a lot of inputs. So, the first time you run it, expect to correct a prompt or two. The point is that the corrections happen in plain chat, not in a timeline. Next, a different kind of workflow, a series. This one is a loop, and on the screen, you have two labels, first video generation on the left and the 100th video generation on the right. Look at the first pair. On the left, a warrior facing a giant creature, and on the

03:38 right, the 100th generation of the same idea from a different camera angle. Same for the street chase and the beach shot at the end. The idea is that when generation is cheap, you do not need to get it right in one go, you run it again, change one line of the prompt, and keep the best one. And here’s a second workflow, a paper craft explainer. It is built from cutout cardboard, tape, hand-drawn arrows, and little microbe drawings in the soil, and you can see the cardboard layer on top. On the left side, you can see Claude Rodder style formula, and it says locked embedded in every scene prompt. So, all

04:10 10 clips in this set keep the same paper craft look. That is the part I find useful for the channel. You decide the style one time, and Claude keeps it consistent across the whole series. Now, let’s go to the website and look at AI filmmaking. This is Cinema Studio 4.0, and the button here says become an AI director. Before I click anything, look at how the camera works here. Every move is a hashtag, static, handheld, zoom, pan. You type it in the prompt or pick it from the library, and you can stack several in one shot, and they run in order.

04:40 Here are product ads made in that style. A burger being assembled in the air with the pickles and tomato slices floating, a set of blue cans tumbling past the camera, a fish eye shot from above. And this is the before and after set. The top row is labeled before, the bottom row is labeled after. Same four ideas, the strawberry pour, the burger, the fish eye character, the chocolate bar. And in the after row, the light, the lens, and the angle are different. There’s also motion control. Look at the small clip of a man’s face in the

05:11 corner. That is the reference movement, and each of the three women repeats his expression. The second variant puts his face on top, so you can compare them side by side. And there’s a personal assistant on the page called Mr. Higgs that can pick the camera, light the scene, write the prompt, or break a script into shots. Seedance and other video models run inside Cinema Studio, and you pick the engine per shot. Let’s stay on the website for a minute, and open the Seedance 2.0 page, because this one shows the workflow in three steps. Step one, input an image reference to guide

05:42 your vision. Step two, write the prompt in natural language, and describe the scene and the sounds. Step three, press generate. So, the sound is part of the prompt. You’re not adding music in another app afterwards. You describe it in the same sentence. Scroll down and the page lists what it is for. Short film and cinematic storytelling campaign ready videos, high impact video ads from product photos, action and effect sequences, and social media content. Two features on this page I want to point at. One-click video recreation where you replicate a

06:14 trending clip and frame-level precision where a character or a font stays locked across the whole video. Mind you, that is what the page promises. The way to know if it holds for your product is to run your own photo through it and you only pay for what you generate. Two more places on the top bar, Canvas and Supercomputer. Canvas first. Canvas is node-based. You chain a prompt node into an image node into a video node and every connection is live. On the page it says cling, soul, and image generation run side by side and outputs route

06:46 between them. It is also shared. The page says your team edits live and every version is saved. So, an ad with five variants becomes one board, not five folders. Now, Supercomputer. This is the agent. Instead of picking a model, you describe the job and it picks from a list of skills. You can see the examples on screen. A cartoon faceless video about a journey to Troy, a UGC ad for a bag of chips with the store page shown on screen, a brand kit for a banana milk, and a whole website built from one product photo.

07:16 And the explainer skill has styles you can pick. Retro 8-bit pixel art, claymation, paper collage, whiteboard doodle, low poly 3D, and plush toy. That is the same idea as the papercraft workflow we saw earlier, but without writing the style formula yourself. Now, the question everyone asks, what does it cost? I opened the pricing page and read the numbers directly. So, these are the real ones. Plans are billed annually. Basic is $9 a month, Pro is 29, and Max is 59. On Max you get 3,600 credits a month, which the

07:47 page says is about 240 seed ends 2.5 videos. Now, per generation, a 5-second Kling 3.0 clip at 720p is about six credits. At 1080p it is about seven. Seed ends 2.0 at 720p is about 22 credits for 5 seconds and about 60 at 1080p. On images, Soul 2.0 is 0.12 credits per image, GPT images one credit, and Nano Banana 2 is two credits. So, if you’re hunting for the cheapest AI platform, compare these per clip

08:18 numbers against what you pay now. For a lot of short social video, six credits for a clip is a small number. And here’s the limit I promised. The page says unlimited models and free generations only work on Higgsfield.ai. They do not work through MCP, CLI, Canvas, or Supercomputer. So, the chat workflow uses credits. So, in this video we saw how to connect Higgsfield to Claude, how Claude generated a vertical video with Kling 3.0 on its own, how a series keeps one style, and what the real credit numbers are. The Higgsfield

08:48 link is at the top of the description. Go ahead, connect it to Claude, and run one prompt off your own. Let me know in the comments what you generated. I’ll see you in the next video.