higgsfield-soul-id

Run higgsfield soul-id create with a name, a model variant, and 5-20 face photos. The CLI uploads the photos, trains the Soul Character, and returns a reference_id you pass to higgsfield-generate with --soul-id for all future generations.

Soul training requires a minimum of 5 photos and accepts up to 20, with 8-12 being the recommended range. Use varied angles, lighting, expressions, and distances, with sharp images of at least 1024x1024 resolution.

The --soul-2 variant is optimized for still image generation and is the default choice. The --soul-cinematic variant is designed for cinematic and video work where motion fidelity matters, so pick based on your downstream use.

No, Soul Character training requires a Basic plan or higher. Check your plan with higgsfield account status before submitting, since the command fails mid-run with a plan error on free accounts.

Training commonly fails due to fewer than 5 photos, heavy occlusion like sunglasses or hats, group photos confusing identity, or uploading videos instead of images. Swap in clearer, more varied photos and retrain.

Use Soul training when you need the same face across many generations or an entire campaign. For a single image with a face reference, use higgsfield-generate with --image instead, and for fictional avatars use a text prompt.