Perfect identity lock is not a realistic promise for generative AI. A better workflow preserves the strongest available references, carries the same character context forward, and gives the table a clear place to decide which results are canonical.
Keep the portrait that best captures the character’s facial structure, ancestry, age, hair, skin, and distinguishing marks. Later generation can use that selected image as the primary identity reference instead of relying only on descriptive text.
A full-body reference can carry armour, clothing, equipment, body proportions, and silhouette into later work. It complements the face reference rather than replacing it.
Names, descriptions, ownership, character type, and selected images live with the registry record. This reduces prompt drift and gives future requests a consistent source of campaign context.
AI output can still change facial details, hands, clothing, scale, composition, or style. Consistency improves through good references and careful selection, not through a guarantee that every generation will be identical. The archive records what your table accepts as part of its visual history.