Seedance 2.5 — Générateur vidéo IA
Le modèle vidéo nouvelle génération de ByteDance : clips de 30 secondes, jusqu’à 50 entrées de référence et retouche par zone. Découvrez-le ici — et générez avec Seedance 2.0 en attendant.
Seedance 2.5 n’est pas encore public — votre prompt s’exécute aujourd’hui sur Seedance 2.0.
Vidéos d’exemple avec prompts complets
Aperçus de Seedance 2.5 — chaque clip est livré avec le prompt exact et les réglages qui l’ont produit. Copiez-les, adaptez-les, appropriez-vous-les.
Plan-séquence 30 s : un ballon à travers 3 000 ans
Produce a 30-second short popular science video about the three thousand years of evolution of football. The entire film uses the same ball as the main visual line. The ball rolls, travels, and deforms from ancient times, connecting different civilizations and eras. The overall rhythm is compact, the graphics are high-end, the historical science short film is combined with artistic transitions to highlight the feeling of a ball spanning three thousand years, and the voiceover is simple and powerful. At the beginning, an ancient ball slowly appeared from the black background with a texture of time on the surface, and then rolled into a Cuju scene from the Warring States Period in China. The picture turned into an ink style, referencing the style of <<<image_1_1>>>. Ancient people dressed in ancient costumes played Cuju in the courtyard with elegant movements and the ball bounced under their feet. Voiceover: The football story starts with Cuju. Then, the ball continues to roll forward, and the picture naturally transitions to an ancient Greek ball game scene. The picture is in the style of a classical oil painting, and the style refers to <<<image_2_2>>>. The background of the square and stone pillars is obvious, and people wearing ancient Greek robes play football. The picture is thick and historical. Voiceover: The Greeks loved ball games too. Then the ball rolled into medieval Europe, and the picture still maintained the style of oil painting. Villages, mud fields, and ordinary people chased the ball. The atmosphere was warm and rough, just like ancient folk football continuing the fire. Voiceover: European folk football keeps the game alive. Then the ball was kicked out, and the screen switched to a black and white documentary style. Referring to <<<image_3_3>>>, the scene arrives in 1863 England. Gentlemen, clubs, and grass pitches gradually appeared, symbolizing the official birth of modern football. This ball showed the standard appearance of modern football for the first time. Voiceover: 1863 — modern football takes shape. Then the scene quickly enters the modern era, with the ball spinning in the air, bringing out key development nodes in turn. The lights, stadium, spectators, trophies, and different scenes around the world are intertwined, showing that football has evolved from a local sport to a global one. At the end, the ball is in the center of a modern stadium, and crowds and cheers from all over the world merge in the background, forming the feeling of "a ball connecting the world". The picture is grand and epic. Voiceover: Now, football connects the whole world.
720p · 31 s · audio natif · 3 images de référence · prompt en traduction anglaise
17 entrées de référence, un seul plan continu
Core instructions: A 26-second one-shot narrative short film, with stable tracking and interweaving, refer to <<<video_1_1>>>, and smooth surround movement, refer to <<<video_2_2>>>. Smooth travel feeling. The alternation of day and night and the flow of the four seasons are realized within the lens. The protagonist is a European woman <<<image_1_3>>>, placed in a sea of people full of fireworks, highlighting the ultimate sense of loneliness and the quality of cinematography. Segmented camera movement and scene description: 0-3 seconds (smooth back): The old wooden door <<<image_2_4>>> opens with a creak, and the camera follows the figure of a European woman wearing <<<image_3_5>>> walking out. She pauses slightly at the threshold. The streets ahead were mottled with light and shadow, and the sounds of hawking and crowds were approaching. She looked distant and slowly walked into the street. 3-6 seconds (back and side tracking shot): The camera keeps a smooth follow-up, she walks into the crowded morning market, the atmosphere refers to <<<video_3_6>>>. The two sides were crowded with colorful fruit stalls and spice shops, and a group of street jugglers were breathing fire dragons, refer to <<<image_4_7>>>. The firelight illuminated the crowd, but she did not squint and walked through at a steady pace. 6-9 seconds (Smooth Surround from the Side): The camera begins to smoothly wrap around to the side and front, capturing her profile. She walked past the noisy butcher's shop <<<image_5_8>>>, and a young mother passes by her holding a baby <<<image_6_9>>>. The baby stared at her curiously, but she just lowered her eyes slightly to avoid looking, without stopping at all. 9-12 seconds (forward and backward follow-up shooting): The camera continues to circle directly in front of the protagonist and performs backward and follow-up shooting. The crowd in front suddenly retreated to both sides naturally like Moses parting the sea. A huge elephant <<<image_7_10>>> covered in gorgeous red cloth appeared from the right side of the screen with a steady pace, occupying most of the screen. 12-15 seconds (gap penetration and rewinding): At the moment when the woman and the elephant are about to collide, the camera cleverly slides through the narrow gap between the elephant and the woman, recirculating back to her back. The elephants passed by hugely and silently, and the urchins chased them with joy. Like bells and laughter, she didn't even slow down. 15-18 seconds (ambient light and shadow gradient): As she walks, the light and shadow in the long shot change magically - the dazzling sunlight in midsummer softens instantly, a breeze rolls up the golden leaves <<<image_8_11>>> in the sky, and the season seamlessly transitions to late autumn in the same long shot. Fallen leaves brushed her shoulders. 18-21 seconds (360-degree immersive surround): The front suddenly falls into a grand street celebration <<<image_9_12>>>. Colorful ribbons and shredded paper burst into the air, and vendors leaned out to cheer. At this moment, the camera unfolds a continuous 360-degree panning movement, creating an extremely strong visual tear between the quiet and lonely protagonist and the frenzied surroundings. 21-24 seconds (circling back to the side and back): When the camera circled around and returned to her side and back, the falling ribbons had quietly turned into snow all over the sky - it was winter in an instant <<<image_10_13>>>. Pedestrians held up umbrellas or put on hoods, and the woman hunches slightly, turns up her coat collar, changes into <<<image_11_14>>>, and continued to walk alone in the snow. 24-26 seconds (slow push and beat): As she walked towards the end of the long street, the sky darkened at a speed visible to the naked eye, and the day sank seamlessly into the night. The dim street lights on both sides and the light bulbs of the stalls were turned on one after another <<<image_12_15>>>. The vendors were packing up their goods. The noise seemed to be slowly absorbed and distant by the heavy snow, and her steps gradually slowed down. Grand fireworks suddenly bloomed in the night sky <<<image_13_16>>>, the sound of fireworks blooming refers to <<<audio_1_17>>>. Colorful light spots flickered and jumped on the building walls and in her eyes. The world is still lively, but she looks up quietly, the camera slowly zooms out, and ends here gently.
720p · 27 s · audio natif · 17 entrées de référence · prompt en traduction anglaise
Retouche par zone : même duel, nouveau château, nouveaux costumes
Replaced the original version of the two-person martial arts video <<<video_1_1>>> with an empty-handed warm-up before a cold-weapon duel. The scene is replaced with a medieval stone castle platform, an ancient courtyard flat, a mountain fortress outer platform, or a simple stone brick duel field. The background is the ancient castle wall, wind, fog, distant mountain line, and the ground is flat and stone <<<image_1_2>>>. The clothes of the man in dark clothes in the video are replaced by <<<image_2_3>>>, and the clothes of the man in light clothes in the video are replaced by <<<image_3_4>>>. The action remains the same, without changing the original rhythm. AI special effects only enhance the environment and texture: wind blown clothes, light fog, a small amount of dust at contact points, metallic cold reflective texture, slight particles and epic color palette. The overall style is restrained, realistic, and a classic hardcore duel atmosphere. Background music synced to the beats.
1112×834 · 8 s · audio natif · 4 entrées de référence · prompt en traduction anglaise
Plan-séquence FPV : bonjour en 11 langues
One-shot FPV drone first-person video, 33 seconds of continuous long shot, no editing, no jump cuts, no transitions. The camera starts from inside the high-altitude clouds and forms a continuous descending flight line along the clouds, fog, light and shadow, valleys, waterfalls, lakes, flower fields, urban buildings and near-ground squares. 11 clear and independent language display blocks appear in sequence throughout the process. Each block only displays the text corresponding to a single language. There is no mixing, overlapping, or adding other languages. 0–3 seconds,<<<image_1_1>>> clouds naturally form “Hello” in Chinese; 3–6 seconds,<<<image_2_2>>> Mist and volumetric light form English “Hello”; 6–9 seconds,<<<image_3_3>>> High-altitude water vapor and sunlight projection form the Spanish (Mexico) "Hola"; 9–12 seconds,<<<image_4_4>>> Ribbon in the sky forming the Indonesian word “Halo”; 12–15 seconds,<<<image_5_5>>> kites formation forming “Hai” in Malay; 15–18 seconds,<<<image_6_6>>> Valley morning mist forms Thai “สวัสดี”; 18–21 seconds <<<image_7_7>>> Waterfall mist forming Arabic مرحبا 21-24 seconds,<<<image_8_8>>> The reflection on the lake and the ripples form the Portuguese word "Olá"; 24–27 seconds,<<<image_9_9>>> Flower fields and meadows are naturally arranged into Vietnamese “Xin chào”; 27–30 seconds,<<<image_10_10>>> City glass buildings reflect light and shadow to form the Japanese word "こんにちは"; 30–33 seconds,<<<image_11_11>>> Nearby fountain water mist, floor paving and light strips form the Korean word "안녕하세요". The overall atmosphere is an early morning sunrise, with golden backlight, soft volumetric light, real clouds and fog, natural motion blur, and movie-level realism. The camera speed starts slowly from 3–5 m/s, gradually accelerates to 14–16 m/s across the natural landscape, and then slows down to 2–3 m/s to hover stably in the near-Earth square. Lens parameters: wide-angle lens, 24fps, smooth FPV drone movement, pitch gradually transitions from -5° to -18°, and finally returns to 0°; slight yaw ±10°, roll controlled at 0–10°, ensuring a continuous, stable, and realistic sense of flight from shot to shot.
720p · 30 s · audio natif · 11 images de référence · prompt en traduction anglaise
Film de marque cinématographique à partir d’un prompt texte
[Overall style] A 30-second high-end brand-level visual blockbuster with a strong cinematic feel and high-end texture. The picture emphasizes dreamy light spots (Bokeh), silky motion blur transitions (Motion blur), volumetric lighting, and ultra-realistic material detail expression. [Storyboard] [0-5 seconds]: Dream prologue and macro close-up Extremely high quality macro close-ups. A slender hand stretched into the air, and its fingertips touched colorful spots as bright and twinkling as stars. With the flow of light and shadow, the scene seamlessly and smoothly transitions to an elegant woman wearing a pure white tulle skirt, who is playing a vintage piano intoxicatedly. Shallow depth of field, the background blurs into a beautiful blue-green tone. [5-15 seconds]: Then cut to a smooth follow shot: a woman wearing a French wide-brimmed straw hat and a flowing white skirt, running lightly in the dense flower path full of pink-orange roses and blue hydrangeas. The light casts dappled light and shadow through the leaves, perfectly showing the gentle breeze, the realistic physics of the skirt fabric fluttering, and the ultra-realistic texture of the petals. [15-24 seconds]: Quiet aesthetics and light and shadow refraction The rhythm of the camera slows down and enters the ultimate beautiful slow motion (Slow-motion). A girl is sitting at a black wrought iron table next to a European retro golden fountain and reading quietly. There are crystal clear soap bubbles floating in the air, and the surface of the bubbles perfectly reflects the surrounding flowers and warm sunshine. The moment when water droplets splash is clearly visible, demonstrating the model's top-level rendering capabilities for transparent materials, water refraction, and complex lighting.
720×960 · 24 s · audio natif · texte en vidéo · prompt en traduction anglaise
Court-métrage anime de course, storyboardé seconde par seconde
30 seconds cinematic youth racing short film, 2d animation style. The protagonist is a young driver who drives a motorcycle to participate in high-profile competitions. The overall style is passionate, youthful, emotionally intense, and cinematic, with a complete beginning, transition, and clear emotional arc. Only two types of camera movements are used in the entire film: high-speed tracking and slow-motion surround. There are very few lines, and they appear naturally like memory fragments. The tone is sincere, gentle, and restrained, without shouting slogans or overly sensationalizing. Don’t feel disaster, don’t express negatively, and don’t exaggerate science fiction. It focuses on love, support, counterattack and growth in the race of youth. 0 seconds to 5 seconds The track begins at dusk with high-speed and intense racing. The camera follows the young man's motorcycle closely to the ground at high speed. The tires skim the edge of the track. The motorcycle roars, the wind blows fiercely, and the atmosphere is tense and fiery. The young man was concentrating, and the setting sun drew sharp highlights on the metal shell of the car. 5 seconds to 9 seconds After entering a key corner, the boy was suddenly overtaken by his opponent. The high-speed tracking continued, and the picture showed the oppressive feeling of the ranking declining and the rhythm being disrupted. In the close-up view of the helmet, there is temporary loss of concentration, tightness of breathing, and slight shaking. The boy whispered: "Can I still catch up..." 9 seconds to 14 seconds The boy fell behind, his breathing became heavier, and his mood hit a low point. The race did not stop and the locomotive was still moving forward at high speed. The scene began to flash back warm memory fragments during high-speed riding: when he was learning to drive as a child, someone supported him from behind; his father arranged his helmet for him, his movements were meticulous and quiet; before the finish line, a gentle smile looked at him; and the back figures walking side by side on the slope at dusk. These memories are presented with golden backlighting, soft slow motion, and fragmented feelings. 14 seconds to 18 seconds The music gradually turns from depressive to uplifting. A restrained and gentle voice came from the memory: "Don't be afraid — I'm always here." "Stay steady. And look forward." The young man's eyes refocused, his breathing slowly stabilized, and his mood changed from wavering to firm. 18 seconds to 23 seconds The young man regained his confidence, accelerated with all his strength, and counterattacked accurately. The high-speed tracking shots show the power and control of the motorcycle when cornering, exiting corners, and approaching the car in front. The boy said quietly but firmly: "I won't stop here." 23 seconds to 27 seconds An ascending track appeared ahead, and the boy sprinted at full speed against the setting sun. The picture only retains the sound of breathing, engine sounds and continuously rising music, without adding unnecessary lines. The locomotive soared into the air with the help of inertia and entered shocking slow motion. A gentle voice with a smile came from deep in my memory: "Go on." 27 seconds to 30 seconds The camera circles the locomotive in mid-air for a slow-motion panoramic close-up. Push the emotions of passion, tenderness, freedom and upward leap to the climax. Flowers bloom behind you, followed by seedance, refer to <<<image_1_1>>>
720p · 30 s · audio natif · 1 image de référence · prompt en traduction anglaise
Qu’est-ce que Seedance 2.5 ?
Seedance 2.5 est la nouvelle génération du modèle vidéo Seedance de ByteDance, dévoilée à la conférence FORCE de Pékin en juin 2026 comme successeur de Seedance 2.0 — le modèle qui domine actuellement l’arène image-vers-vidéo d’Artificial Analysis.
À l’heure où nous écrivons, Seedance 2.5 n’est pas publiquement disponible. La page Dreamina de ByteDance l’affiche toujours « coming soon », et l’accès se limite à des tests d’entreprise fermés. Tout site tiers prétendant vous faire générer avec Seedance 2.5 aujourd’hui fait tourner un autre modèle en coulisses.
Une fois lancé, ByteDance a indiqué que Seedance 2.5 propulsera ses applications grand public — Jimeng et Doubao en Chine, Dreamina et CapCut à l’international.
Nouveautés par rapport à Seedance 2.0
- Clips de 30 secondes en un seul plan, contre 15 auparavant — avec un mode vidéo longue en bêta qui va jusqu’à 180 secondes (annonce officielle).
- Jusqu’à 50 entrées de référence multimodales — texte, images, vidéo, audio et style — contre la limite de 9 images, 3 vidéos et 3 clips audio de Seedance 2.0 (annonce officielle).
- Retouche par zone : modifier une région précise d’une vidéo générée sans re-rendre tout le clip (annonce officielle).
Essayer en trois étapes
Ouvrir le playground
Seedance 2.5 est encore en test fermé : le playground tourne donc sur Seedance 2.0 — même famille, disponible dès maintenant. Les nouveaux comptes reçoivent 10 crédits gratuits.
Écrire ou coller un prompt
Partez d’un prompt d’exemple de cette page ou décrivez votre propre scène : sujet, action, caméra, style.
Générer et itérer
Examinez le résultat, ajustez votre prompt et relancez. Nous basculerons cette page sur Seedance 2.5 dès son ouverture.
Capacités et limites
Ce que ByteDance annonce
- Clips de 30 secondes en une génération, extensibles à 180 secondes via un mode vidéo longue en bêta (affirmation officielle).
- Jusqu’à 50 entrées de référence — texte, image, vidéo et audio — pour la cohérence des personnages et du style (affirmation officielle).
- Retouche par zone sans régénérer le clip entier (affirmation officielle).
- L’audio comme entrée de référence prise en charge, dans la lignée de la génération audio synchronisée native de Seedance 2.0 (affirmation officielle).
À savoir avant de planifier un projet
- Pas encore public : aucun accès via les applications grand public. Les sites tiers vantant un accès immédiat à Seedance 2.5 utilisent d’autres modèles.
- Les premiers testeurs en préversion rapportent pour l’instant une sortie 720p, alors que les documents officiels promettent la 4K — traitez les chiffres de résolution avec prudence jusqu’à la sortie large.
- Plusieurs spécifications clés, dont la fréquence d’images, restent officiellement non annoncées.
- La politique de référence de visages de Seedance 2.5 n’est pas annoncée — la prise en charge de visages réels sur cette page repose actuellement sur Seedance 2.0.
Vous intégrez la génération vidéo dans votre propre produit ? La page développeur suit le statut d’intégration de Seedance 2.5 et tout l’aspect technique.
API Seedance 2.5 →Questions fréquentes
Everything you need to know about the product and billing.
Explorer la famille Seedance
Dernière mise à jour: