higgsfield-soul-id

Yes, you can maintain a consistent identity across different prompts by training a reusable Soul identity model from face photos. Once trained, you apply the returned reference ID to your generation tasks, ensuring the same identity appears consistently in both image and video outputs.

When training a Soul identity model, you need 5 to 20 varied, well-lit face images of the target person. You also need to select the appropriate training variant depending on whether you want to use the identity for default image generation or cinematic video use.

No, you do not need to retrain the Soul identity model for each new prompt. The trained model is reusable, allowing you to apply the same identity reference across multiple generation tasks and prompts for long-term use.

A Soul identity model is a reusable representation trained from a person's face photos that enables identity-faithful image and video generation. You need it when creating a digital twin, generating avatars, or maintaining the same identity across multiple generation tasks and prompts.