Write a compact character brief
Choose an original fictional adult and record a small set of stable traits: age, face shape, hair, eye color, distinguishing details, wardrobe, and visual style. Keep scene instructions such as location or camera motion separate so you can change a shot without rewriting the identity.
For a realistic AI character generator, coherent photography instructions help: choose one lighting setup, a clear view of the face, and a background that does not compete with the subject. The prompt below is a starting recipe, not a tested guarantee of likeness or quality.
A reusable front-facing adult character portrait
Photorealistic studio portrait of an original fictional 32-year-old woman with an oval face, shoulder-length dark auburn hair, brown eyes, and a small freckle beside her left cheek. Wearing a fully buttoned ivory blouse. Face, shoulders, and torso directly toward the camera, symmetrical relaxed posture, eyes at camera level. One person, both shoulders visible, soft even studio lighting, plain warm-gray background, natural skin texture, no text or logos.
Copy this recipe, then paste it into the matching prompt field.Approve and preserve the reference image
Review the eyes, teeth, ears, hairline, hands, and clothing before using a portrait downstream. Keep the original downloaded file together with its prompt, model, settings, and task ID. Avoid repeatedly compressing or retouching the reference between steps.
If you are creating an AI virtual model for several wardrobe concepts, keep a separate record for each approved look. A fresh text-to-image generation may reinterpret the face. Use image references in a compatible workflow when visual identity matters, and review the result rather than assuming it is preserved.
- Keep one approved reference as the identity source.
- Use a neutral, unobstructed face for talking or face-swap tasks.
- Record which image, source clip, and generated output belong together.
Choose animation, face swap, or a talking avatar
For a moving version of the still, use Text & Image to Video in Playground and supply the portrait as a reference. The Minimax H3 Workshop itself takes text; writing the same character description there is not the same as passing the approved image.
For a clip whose motion and framing already work, Video Face Swap takes the source video and an authorized reference face. It replaces a face rather than guaranteeing the character's whole body, clothing, or environment will match the portrait.
For a scripted presenter, use Workshop Avatar or Lip Sync. In Avatar, an existing saved portrait with a linked base video can reuse that video for the lip-sync stage. A new portrait without a linked video still needs the corresponding preparation steps.
Review continuity through the entire clip
Start with a short shot. Check the first and last frames, then watch through expressions, head turns, and occlusions. Compare the face with the reference and look for changes to the outfit, background, camera framing, and body orientation.
If a reusable talking base is your goal, request a locked camera, the same pose as the reference, and a constant background. These instructions guide the model; they do not enforce zero motion. Keep the original base video so you can reuse a reviewed result rather than generating a different performance every time.
Frequently asked questions
Will the same seed guarantee the same AI character?
No. A seed may help reproduce a particular request where supported, but it is not an identity guarantee across changed prompts, models, images, or video tasks. Keep and reuse an approved reference image.
Does face swap preserve the whole character?
Face swap replaces a face in an existing video. It does not by itself reproduce the reference portrait's body, wardrobe, scene, or performance. Choose a suitable source clip and review the whole output.
Can I reuse a saved avatar without generating its base video again?
When a saved avatar has a linked completed base video, selecting that asset in Workshop Avatar lets the workflow reuse it for lip sync. A standalone image without a linked base video still needs video preparation.