One of Higgsfield MCP's most useful features is Soul character training — train a character once and reuse it across every scene. Here is how it works through the connector.
What Soul training does
You upload a handful of reference photos, train a reusable 'Soul' character, and then reference that character in generations. Identity stays stable across renders — the failure mode that usually breaks AI video series.
- Train once from reference photos.
- Reuse the same character across images, videos, and brand moments.
- Combine with voice cloning and multilingual lip sync.
How to invoke it via the agent
Through the MCP connection, you can create and list trained characters and ask the agent to use a specific character in a generation. The connector exposes character management alongside generation.
- show_characters — list what you've trained.
- Reference a character by name in your prompt.
- Iterate without re-uploading references each time.
Best practices
Use 20+ varied, high-quality reference photos with a cohesive style and no faces if you want faceless consistency. Avoid blurry or mixed-style inputs.