Digital Human Identity Studio

What It Solves

When generating digital human images, the harder problem is not one shot but many shots that keep the same identity. If hair, eyes, accessories, and age impression are rewritten ad hoc, the character drifts. This skill separates identity from scene: a fixed Character Sheet defines “who they are”, while a variable Scene Card controls “what they are doing”.

How It Works

• Identity initialization: store hair color, eye shape, signature_accessory, core_vibe, and style in character-sheet.yaml, then mark meta.confirmed: true; scene generation does not start without a confirmed identity.
• Scene generation: call MiniMax image-01 via the image_generation endpoint, render prompts with gen.py, and download data.image_urls[0].
• Consistency anchoring: optional image_file references can lock face and body; the docs warn that image_base64 is broken on MiniMax.
• Audit fallback: for rejections such as 1033/1026, retry up to twice by softening outfit and pose in the Scene Card, never by changing the Character Sheet.

Boundaries

It suits engineers and creators building repeated character visuals, photo sets, or Cosplay images. It avoids explicit, lingerie, or body-keyword workflows. The safety rule is to describe clothing rather than body, such as form-fitting dress that hugs her curves instead of direct body terms.