Capability

Higgsfield — generate images & video

AI image and video generation in Claude via the Higgsfield MCP — 30+ cinematic models behind one connector.

#mcp#video#image

The Higgsfield MCP brings AI image and video generation into Claude. Connect it once and, from Cowork or any chat, you describe the shot you want — Claude picks the right model, sets the parameters, fires the generation, and brings the finished image or clip back into the conversation. It exposes 30+ models behind one connector.

Describe a shota prompt
Higgsfield MCPpicks model + params
Generate30+ models
Back to chatimage / video
One connector, every Higgsfield model — describe it and Claude handles the rest.

Connect it

Higgsfield is a remote (hosted) connector — perfect for Cowork and the Claude app. Setup is one-time.

  1. 1

    Open Connectors

    In Cowork / claude.ai (web or desktop): Settings → Connectors, then Add custom connector.
  2. 2

    Paste the server URL

    Name it Higgsfield and paste the MCP server URL:
    text
    https://mcp.higgsfield.ai/mcp
  3. 3

    Connect & sign in

    Click Connect; Claude redirects you to sign in to your Higgsfield account. Approve access and the connector stays active.

Claude Code & other CLI agents

For Claude Code, Codex, and similar, Higgsfield recommends its CLI or Skills instead of the remote connector.

What it can do

Cinematic image-to-video

Turn a still into a moving, directed shot.

Marketing video generator

Produce ready-to-post promo clips from a brief.

Viral clip generator

Generate short-form clips tuned for social.

Soul character training

Train a consistent character to reuse across generations.

Video analyzer

Analyze a video and pull out what makes it work.

Virality prediction

Score how likely a clip is to perform.

Plus the basics across every session: generate images, create videos, train consistent characters with Soul, and browse your generation history to reuse as input.

Models & limits

A single connector fronts 30+ models — Claude chooses per request. A sample:

SoulCinema StudioKling 3.0Seedance 2.0Veo 3.1Nano Banana ProGPT Image 2Seedream 4.5
4K

Image resolution, up to 4K.

15s

Video length, up to 15 seconds, in many cinematic styles.

Soul

Train a character once, keep it consistent across shots.

Generate something

  1. 1

    Describe the shot

    Be specific about subject, style, motion, and length.
    text
    Make a 10-second cinematic product shot of a matte-black watch
rotating on a pedestal, soft studio light, shallow depth of field.
  2. 2

    Let Claude pick the model

    It maps your intent to the right Higgsfield model and parameters — no model menu to learn.
  3. 3

    Collect the result

    The finished image or clip comes back into the chat; iterate with follow-ups, or train a Soul character to keep a subject consistent across generations.

Programmatic video too

Higgsfield is generative (describe → model renders a clip). For deterministic, HTML-native video you compose and render yourself, see HyperFrames.

Learn more