绫波丽双次吐舌表情深度驱动 · MiniMax H3
15 秒真人风短片，怎么做出来的？

subject_definitions:
<Subject 1> is Rei in <Picture 1>: the exact same face, natural skin, dark eyes, straight long black hair with blunt bangs, light-blue crewneck knitted sweater and silver necklace.
<Subject 2> is the white textured wall and sofa edge in <Picture 1>.
<Picture 1> defines identity, clothing, background, lighting, framing and the opening image.
<Video 1> is a diagnostic depth animation with facial motion curves and a tongue silhouette. Read it only as a guide to movements and their timing; all visible surfaces, colors, facial proportions and illumination come exclusively from <Picture 1>.

summary:
[reference generation] A continuous fifteen-second photorealistic medium close-up of <Subject 1>, fixed eye-level camera, landscape 16:9. Keep the same face, black hair, light-blue sweater, necklace and white wall throughout. She makes two brief playful tongue-out gestures at the original reference times, retracts her tongue after each and continues subtle facial expressions. No speech or cuts.

retention_analysis:
Fully preserve <Picture 1> appearance and environment. Change only facial expression and small head movements. Preserve natural shoulders and continuous knitted sleeves. Translate the motion into natural skin and knitted cloth. The visible result is a normally illuminated color photograph of Rei, with her original jaw width, cheek volume and chin length.

detailed_description:
[Shot 1]
Begin from <Picture 1>, then smoothly perform the expression. From 0.00 to 3.20 seconds, perform the initial eye, lip and small head movements with the tongue inside the mouth. Around 3.20 seconds the lips part and a naturally pink tongue extends over the lower lip; it retracts around 4.07 seconds. Continue the intervening expressions and small gestures. Around 8.81 to 10.68 seconds, perform the second tongue extension, brief hold and retraction. From 10.68 to 15.00 seconds, continue the remaining eye, mouth, head and hand gestures at the original pace. Two tongue cycles only, naturally connected inside the mouth. No looping, speed change, reset or forced freeze at the end. Keep the face recognizable and skin realistic. Keep the camera fixed and the original face scale. Do not change the light-blue sweater or white wall. No gray depth-map rendering, colored facial curves, dots, masks, text or graphics. No detached tongue or distorted arms. Original audio is added separately; do not generate speech.