higgsfield-soul-id

Run higgsfield soul-id create with a name, a variant flag like --soul-2, and 5-20 face photos via repeated --image flags. The CLI uploads local paths automatically and returns a reference ID once training completes.

Use 8-12 sharp photos of a single person with varied angles, lighting, expressions, and distances. Avoid group photos, sunglasses, heavy filters, hats, and repeated poses, as these reduce identity capture quality.

--soul-2 is the default variant intended for image generation, while --soul-cinematic targets cinematic and video work. Choose based on your stated downstream use before submitting the training job.

Training fails when there are fewer than 5 photos, heavy occlusion like sunglasses or hats, group photos confusing identity, or non-image uploads. Swap in better varied photos and retrain.

Yes, Soul training requires at least the Basic paid plan. If higgsfield account status shows a free plan, the CLI returns a Minimum Basic plan required error and you must upgrade first.

Do not use Soul training for one-shot face swaps, which higgsfield-generate handles with --image, or for named non-photo characters, which a text prompt handles. Soul is for reusable identity-faithful models.