Realistic Crying Micro-Expressions

Seedance 2.5 + MiniMax H3RealisticImage to VideoLip SyncSeedance 2.5MiniMax H3Handheld CameraOcean3s
A three-part crying scene — holding back, breaking down, sobbing — completed in 3 seconds. Expression instructions with the precision of "tears welling up in the eyes but not yet falling" represent the ceiling test for digital-human acting. Runs on both Seedance 2.5 and MiniMax H3.

PROMPT · Full Video Prompt

Create with AI
Use the reference image as the starting frame. Maintain complete character consistency: a young East Asian woman, grey sleeveless top, large thin silver hoop earrings, soft overcast lighting, a fully blurred coastal background. Maintain a subtle, continuous handheld camera drift. Throughout, her hair is gently blown by the wind. Action sequence: The subject holds back tears. Her lips press tightly together, her chin trembles, her chest rises and falls with sharp, quick breaths. Tears well up in her eyes but do not fall. The emotion breaks. Her eyes squeeze shut, her mouth opens into a sob. Tears actively roll down both cheeks. Her head tilts slightly downward. The front collar of her grey top is pulled straight upward from the bottom edge of the frame, the fabric naturally stretching and deforming as it covers her chin, mouth, and nose. The fabric deforms naturally over her facial features. No hands, fingers, or arms may appear in the frame. The fabric naturally loosens and falls back down, returning to her collarbone. Her face is revealed again, her eyes slightly open and looking downward, her mouth open, breathing heavily and wearily.

✍️ Editor’s Notes

This one is basically a minimalist template that barely needs expanding — the four action beats already break 'holding it in' down to the muscle level: lips pressed shut, chin trembling, chest tight, tears welling but not falling — a sister piece to the earlier collar-covering-the-face short prompt, same technique, which tells you this is a template the author has verified and repeated. It's flagged need_expanded probably because, while the short passage is information-dense, it's missing the environmental detail and camera-breathing texture that both Seedance 2.5 and MiniMax H3 need fed to them. What the expanded version needs to do isn't add new actions — it's feed in the atmosphere layer: handheld drift, hair moving in the wind, skin texture — while keeping the single most critical negative constraint, “no hands may appear in frame”, exactly as-is.

Want tighter control? Try this expanded version:
Opening frame carries over from the reference image: young East Asian woman, gray sleeveless top, large thin silver hoop earrings, soft overcast light, background a fully defocused coastline, skin retaining natural pores and slight asymmetry. Camera holds a continuous handheld drift, amplitude very small, strands of hair blown by the sea breeze one at a time. Action unfolds in four beats: first, intense suppression — lips press into a straight line, chin trembles faintly, chest tightens from a sharp inhale, eyes well up with tears but they don't fall; then the dam breaks — both eyes squeeze shut hard, the corners of the mouth pull downward into a sobbing shape, two lines of tears slide down the cheekbones at the same time, head and neck dip a few degrees; immediately after, the gray collar is pulled upward from the bottom edge of the frame as if by an invisible force, the fabric stretching and deforming realistically against chin, lips, and the bridge of the nose — no fingers, hands, or arms may enter the frame at any point, the occlusion must be achieved purely through the fabric's own tension; finally the fabric releases and slides back down to the collarbone, the face is revealed again, eyes half-open looking down, mouth slightly parted, breathing heavy and exhausted.
Create with AI

Related Works

׋›
↑