Wan 2.6 : vidéos IA de 15s avec cohérence cinématographique et synchronisation labiale parfaite.
Le modèle vidéo de nouvelle génération d'Alibaba — prompts plus intelligents, synchronisation audio améliorée et cohérence de personnage inégalée.
Essayer
Fonctionnalités clés
Génération de Récit Multi-Plans
La plupart des modèles vidéo open-source génèrent un seul clip continu, souvent dépourvu de structure ou de cohérence. WAN 2.6 introduit une percée majeure avec sa capacité à générer des récits multi-plans directement à partir de prompts simples.
Prompt
The scene unfolds in first-person POV inside a bright, refined modern kitchen. Natural daylight pours across walnut flooring and matte gray cabinetry, giving the space a calm and polished atmosphere. The viewer takes three to four slow, steady steps forward while holding an empty celadon-green porcelain bowl with both black-gloved hands. Ahead stands a built-in double-door refrigerator. The left door features a softly glowing dispenser slot, with faint vapor curling from its edges. When the viewer reaches the refrigerator and lifts the bowl beneath the outlet, a gentle mechanical hum begins. From the small dispenser opening, the plating sequence unfolds with precise, almost ritualistic elegance. First, a smooth stream of deep orange lobster bisque flows into the bowl, circling and rippling as it settles. Moments later, tender pieces of lobster claw and tail meat descend into the center, their pink-red surfaces glistening in the hot broth. A thin ribbon of cream follows, tracing a delicate spiral across the bisque. Finally, micro herbs and tiny gold flakes drift down, completing the dish with a soft visual flourish. The celadon glaze of the bowl reflects the bright natural light, while the warm tones of the bisque shimmer gently on the surface. Subtle sounds fill the space: soft footsteps on the wooden floor, the quiet friction of gloves against the bowl, the rising hum of the refrigerator, the thick pour of bisque hitting the ceramic, the gentle plop of lobster pieces, the light drizzle of cream, and the faint sprinkle of herbs and flakes. Altogether, the moment blends mechanical precision with the warmth and intimacy of fine dining, presented through the calm rhythm of first-person ASMR realism.
Final outcome
Génération Vidéo Basée sur Référence
WAN 2.6 prend en charge la génération avec référence vidéo, permettant aux utilisateurs de guider le modèle avec une vidéo d'entrée.
Prompt
character1 is eating dinner with character2 in a restaurant
Final outcome
Génération de Vidéo Longue de 15 Secondes
De nombreux modèles open-source sont limités à produire des vidéos très courtes, généralement seulement 2–5 secondes, ce qui restreint la profondeur narrative. WAN 2.6 brise cette barrière en prenant en charge des vidéos jusqu'à 15 secondes de durée.
Prompt
Generate an approximately 15-second cohesive narrative video. Story: A medieval knight awakens on a storm-swept meadow after a fierce battle. First 5 seconds: A slow circling shot reveals his mud-covered armor, scattered debris, and lingering flashes of lightning in the dark sky. Middle 5 seconds: The knight rises, grasping a sword embedded in the ground. The camera pulls upward from a low angle, emphasizing the determination in his eyes. Final 5 seconds: He begins running toward a distant ruined castle wall as the camera follows in a handheld-style tracking motion, tall grass brushing past the lens to create dynamic depth of field. Maintain scene continuity, natural body motion, and cinematic epic atmosphere throughout.
Final outcome
