Cómo escribir prompts para Seedance 2.0

¿Alguna vez has escrito un prompt en una herramienta de generación de vídeo con IA y el resultado no tenía nada que ver con lo que imaginabas? No eres el único. Esa distancia entre lo que visualizas y lo que genera el modelo casi siempre se debe a lo mismo: cómo escribes el prompt.

Seedance 2.0, el último modelo de generación de vídeo con IA de ByteDance, acorta esa brecha. Es capaz de interpretar tus instrucciones muy bien, como si fuera un director que lee un guión: ángulos de cámara, movimientos de personajes, iluminación, ritmo e incluso audio sincronizado responden directamente a cómo estructures el prompt.

El modelo acepta texto, imágenes, vídeos y archivos de audio como referencias. Puedes combinar hasta 14 elementos en una sola generación gracias a su sistema de referencias con etiquetas @.

Una herramienta tan potente solo funciona si sabes cómo guiarla, así que aquí te explicamos cómo escribir prompts que saquen el máximo partido a Seedance 2.0.

Cómo escribir prompts efectivos para Seedance 2.0

Con Seedance 2.0 hay que cambiar el chip. Ya no estás describiendo una escena, la estás dirigiendo. Esto significa que tu prompt no puede ser una imagen mental vaga; debe tener una intención clara.

Hay cuatro cosas básicas que conviene tener claras. La primera es la longitud del prompt: entre 100 y 260 palabras funciona bien. Con menos te quedas corto y el resultado es genérico. Si es más largo,el modelo puede pierde.

La segunda es el movimiento y aquí hay que ser muy concreto. El modelo no sabe cuánta energía tiene una acción si no se lo aclaras: “un coche que pasa” puede significar cualquier cosa. En cambio, “un coche cruza el plano de izquierda a derecha a toda velocidad, levantando polvo” no deja lugar a dudas. Usar adverbios y expresiones como rápidamente, suavemente, frenéticamente o de golpe parece un detalle sin importancia, pero el resultado cambia bastante.

La tercera es el estilo. Incluye al menos una palabra que ancle la dirección visual: cinemático, documental, comercial, anime, fotorrealista. Sin eso, el modelo tira por donde quiere.

Y, en último lugar, aplica los cambios uno en uno. Si la acción te funciona pero el encuadre no, modifica solo la descripción de la cámara. Cambiarlo todo a la vez hace imposible saber qué fallaba.

Estructura estándar de un prompt en Seedance 2.0

La comunidad lleva tiempo dándole vueltas y hay una estructura de prompt que gana por goleada:

Sujeto + Acción + Cámara + Estilo + Restricciones

• El sujeto define quién o qué aparece en la escena. Cuanto más específico, mejor: edad, ropa, expresión y postura ayudan al modelo a construir una imagen más nítida.
• La acción describe lo que ocurre. Usa el presente y céntrate en un único movimiento principal por plano.
• La cámara le indica al modelo cómo está encuadrada la escena (plano general, plano medio, primer plano) y cómo se mueve: travelling, panorámica, órbita, cámara en mano.
• El estilo define el tono visual: la iluminación, el etalonaje, las referencias cinematográficas o la atmósfera general.
• Las restricciones reducen los errores y mantienen la coherencia. Por ejemplo: “sin distorsión”, “mantén la consistencia facial” o “movimiento estable”.

Si usas referencias, Seedance 2.0 les asigna etiquetas del tipo @Image1, @Video1 o @Audio1. Puedes mencionarlas directamente en el prompt. Por ejemplo:

“Character from @Image1 walks through the landscape in @Image2, camera tracking shot, beat-synced to @Audio1.”

Esto mejora la coherencia y evita que los elementos visuales queden a la deriva entre planos.

Usar marcas de tiempo para un control preciso del ritmo

Cuando tu escena presenta varios momentos bien diferenciados, las marcas de tiempo te proporcionan un control aún mayor sobre la estructura y el ritmo del vídeo.

En lugar de describirlo todo en un bloque continuo, divide la secuencia en segmentos con tiempo asignado. Así, el modelo sabe exactamente cuándo debe ocurrir cada acción, cada movimiento de cámara o cada cambio visual.

Por ejemplo:

[00:00-00:04] A woman stands alone in a foggy street at night. Wide shot. Soft blue lighting.
[00:04-00:07] She begins walking toward the camera. Slow dolly in.
[00:07-00:10] Close-up as she stops and looks directly into the lens. Subtle wind in her hair.

Este enfoque funciona muy bien en:

• Secuencias de múltiples planos.
• Construcciones narrativas progresivas.
• Escenas de acción con una coreografía definida.
• Transiciones de estilo dentro de un mismo clip.

Las marcas de tiempo eliminan la ambigüedad. El modelo no tiene que adivinar qué ocurre primero ni cuánto dura cada momento. Estás definiendo el ritmo de la escena. Es casi como editar en papel antes de generar.

Una cosa importante: cada segmento debe incluir solamente una acción. Cuanto más limpio sea cada bloque, más fluido será el resultado.

Cómo evitar que los elementos cambien entre planos

Uno de los problemas más habituales en los vídeos generados con IA es que algo importante (una chaqueta, un producto, un objeto concreto) aparece ligeramente diferente de un plano a otro. Y cuando eso pasa, toda la secuencia se resiente.

Para evitarlo, la clave es definir un elemento de referencia fijo: algo que el modelo sepa que no puede modificar ni eliminar. Puede ser una prenda, un producto con detalles de diseño específicos, un accesorio, un vehículo o cualquier prop que necesites que se mantenga igual de principio a fin.

Cuando uses imágenes de referencia, sé explícito sobre para qué sirve cada una:

• “Use @Image1 for the jacket design. Keep it unchanged in all shots.”
• “Use @Image2 for the product shape and details only.”

Y si tu secuencia tiene marcas de tiempo o cambios de cámara, recuérdalo en cada bloque. El modelo no da por sentado nada, así que cuanto más lo refuerces, menos deriva visual habrá.

También ayuda rematar el prompt con una restricción corta y directa:

• “Mantén el diseño del producto durante todo el vídeo.”
• “El conjunto no debe cambiar entre planos.”
• “Conserva las proporciones del objeto.”

Los mejores prompts para Seedance 2.0

Lo que tienen en común los prompts que funcionan es que son concretos: dicen qué pasa, cómo se ve y desde dónde se mira. Sin vaguedades.

Aquí tienes ejemplos por categoría. Cópialos, cambia lo que necesites y úsalos como punto de partida.

Prompts para vídeos dramáticos y con narrativa cinematográfica

En este tipo de prompts, lo que marca la diferencia es la especificidad emocional y saber exactamente dónde está la cámara en cada momento. Piensa en cómo se sentiría la escena vista desde una butaca en el cine:

“15-second cinematic documentary trailer, shot on 16mm film, handheld camera, natural light. Follows modern real witches: men and women of all ages with traditional witch aesthetics mixed with contemporary life (black dresses, pointy hats with puffer jackets, sneakers; potions, herbs, crystals next to microwaves, wifi routers, TVs). Sequence: 1. Wide dawn shot of suburban street, one house with smoke from windows and a black cat on the mailbox. Text: “they live among us.” 2. Middle-aged witch in a pointy hat stirring a cauldron in a modern kitchen next to boiling pasta. 3. Young witch on bus scrolling phone, broomstick leaning like an umbrella, passengers glancing. Text: “they have jobs.” 4. A witch in a blazer and a pointy hat giving a corporate presentation, coworkers nodding. 5. Elderly witch watering garden, plants grow fast, she shrugs at the camera. 6. Handheld close-up: hands grinding herbs, then typing on a laptop. Text: “they pay taxes.” 7. Two witches are arguing at the farmers’ market over dried sage, crowd is gathering. 8. Young male witch struggling to parallel park a car with a broomstick on the roof rack. 9. Wide aerial shot rising over neighborhood, houses with subtle witch signs (smoke, cats on roofs, glowing windows, herb gardens, broomstick on porch). Text: “the craft. coming soon.” Fade to black. Soundtrack: slow, melancholic cello melody with subtle comedic notes for funny moments. Muted color palette: overcast suburban greys, warm interior ambers, deep greens, black clothing against pastel houses. Tone: serious, intimate, observational with quiet absurd comedy, photorealistic high production value.”

Prompts para vídeos de animación

Seedance 2.0 maneja bien la animación estilizada, pero necesita que le indiques el estilo con claridad desde el principio:

“15 seconds hand-drawn 2D cel animation opening sequence. Protagonist: young man with spiky black hair in a messy bun, pale skin, dark decorative face paint lines under eyes, all-black high-collar long coat, black boots, black gloves — cool, mysterious look but constantly in silly situations. Sequence: 1. Dramatic close-up of his face in moody green and purple lighting, a bird lands on his head, a serious expression breaks. 2. Cut to title «KOROKORO» in bouncy handwritten letters with cute doodles. 3. Fast-paced comedy montage: him walking dramatically, then stepping on a banana peel and falling; sitting cool at a cafe while stray cats pile onto lap, cannot move; posing on rooftop at sunset then phone rings with cute ringtone; dragged shopping by little old lady who thinks he’s her grandson; getting broth splashed on black coat while eating ramen. 4. Final shot: him on a park bench covered in birds and cats, a flower in his hair, staring at the camera unamused. Freeze frame, title reappears. Lighting contrast: dark, moody lighting for dramatic scenes, bright pastel lighting for comedy moments. Upbeat pop-rock soundtrack.”

Prompts para vídeos de moda

En el ámbito de la moda, lo que manda son el movimiento de las telas, la luz y la composición. Si eso está bien definido, el resto se sostiene solo:

“15 seconds mixed media fashion film combining real filmed footage with animated graphic elements, hand-drawn illustrations, and bold paper cutout shapes layered on top. Dynamic fast-paced rhythmic editing. Group of female models in eclectic vintage fashion (oversized leather jackets, patterned silk scarves, wide-leg trousers, chunky platform boots, layered necklaces, round sunglasses) filmed in real locations: laundromat (lavender color grade), parking lot (orange), diner booth (mint green), stairwell (hot pink). Animated graphic overlays react to models: bold geometric shapes in coral, violet, lemon yellow slide in like paper slapping the screen, partially covering/revealing models. Crude hand-drawn illustrations animate in real time: wiggly, blinking eyes; spinning planets; arrows pointing to outfit details; zigzag lines radiating from the spinning model; a bouncing, drawn crown. Fast shot sequence: model blows bubblegum in diner → cut to model leaning on washing machine with drawn soap bubbles → cut to hand pulling sunglasses down revealing cartoon eyes → cut to two models walking in sync down stairwell from above with paper cutout butterflies → cut to model kicking platform boots toward camera with drawn impact star. Split screen: four models in four locations simultaneously, each in a different background color, moving to the same beat. Halftone texture flashes over shots, turning into high-contrast risograph-style two-tone prints (violet & cream, coral & black). Handwritten words/symbols/arrows scribble across the frame between cuts in thick marker. Final wide shot: all models together in laundromat, frozen mid-laugh, while animated confetti, shapes, doodles, and text explode across the frame and hold. Aesthetic: raw, joyful, loud zine culture meets runway meets cartoon chaos. Color palette: lavender, orange, mint green, hot pink, coral, violet, lemon yellow, black, cream. High energy, photorealistic footage mixed with flat graphic animation.”

Prompts para vídeos con mucho movimiento

Si tu objetivo principal es generar vídeos con movimiento dinámico, apóyate en verbos de acción potentes y en señales de ritmo. Funciona muy bien para movimiento atlético, mecánico y natural:

“Motion design video, 15 seconds, ultra sleek, minimal aesthetic, black background throughout. The sequence begins with a completely black screen. A single white point of light appears center frame. It pulses once — a soft circular wave ripples outward like a heartbeat. With each pulse, the point emits expanding concentric rings that morph into increasingly complex forms in a continuous, unbroken evolution. First pulse — the point stretches into a perfect thin white line that extends horizontally across the frame. Second pulse — the line multiplies into a grid of parallel lines that rotate and form a geometric circle. Third pulse — the circle extrudes into a 3D wireframe sphere that spins slowly, every vertex and edge crisp and precise. The wireframe fills in smoothly, becoming a solid white surface with subtle light reflections. The sphere flattens and unfolds like origami into a rectangular plane that becomes a minimal phone interface — with clean lines and abstract UI elements. The interface folds inward, collapsing into a small cube that tumbles once and opens like a blooming flower — each petal a smooth white surface that peels outward. The flower form dissolves into hundreds of tiny particles that swirl in a controlled vortex, each particle following a precise mathematical path. The particles slow and reorganize — converging, aligning, compressing — forming the shape of a wristwatch seen from the side, every detail built from the same particles. The watch dissolves again, the particles stream across the frame in fluid ribbons that weave together and compress into a single dense white sphere. The sphere pulses one final time — brighter than before — and collapses inward to a single point. The point holds for one beat in silence. Then it smoothly expands into the brand logo — clean, minimal, white on black — with a single subtle accent of warm amber light that glows once behind it and fades. Every transition between forms is seamless, fluid, and continuous — no cuts, no jumps, one single.”

Prompts para videoclips

Este tipo de prompt debe hacer referencia al ritmo, al ritmo de edición y a la energía visual. Si tienes un archivo de audio, usa la etiqueta @Audio para sincronizar el movimiento con el beat:

“15 seconds mixed media collage animation music video, stop-motion-like movements, jerky transitions, constant shifting layered compositions. Young woman singer and band in fragmented collage form: mismatched photographic cutouts on painted backgrounds, white torn-edge outlines. [0:00-0:03] Close-up of singer’s face: misaligned photographic fragments (eye from magazine, lips from another, hand-painted black hair), vibrating on layered torn posters/newspaper/halftone background with neon pink arrows. [0:03-0:06] Pull back to full figure: photorealistic face on painted golden checkered dress, in a flat room with crooked lamp and melting clock, floating torn paper and Japanese typography. [0:06-0:09] Band playing: headless suit cutouts with pulsing colored silhouettes (red/blue/green), hand-drawn ink instruments, background swaps on beat (saturated blue with splashes, lavender geometric shapes, neon green halftone). [0:09-0:12] Quick cuts: photographic hand snapping fingers, extreme close-up singing lips with concentric circles, sneaker walking from torn magazine pages, transparent colored shapes sliding across the frame. [0:12-0:15] All collage elements swirl into the center, compress into the band logo, flash white. Handmade tactile imperfect style: visible glue marks, torn edges, registration errors, and ink smudges. Color palette: bold primary blocks (electric blue, vermillion red, white), muted mid-century tones (lavender, gold), neon accents (pink, acid yellow, lime green) with halftone textures. Punchy rhythmic energy, every cut synced to music.”

Prompts para vídeos comerciales y de producto

Los prompts centrados en producto requieren una composición limpia, iluminación controlada y movimiento mecánico fluido:

“Premium headphone campaign video, 15 seconds, single unbroken continuous shot, no cuts, smooth Steadicam movement flowing through connected spaces. The product is a pair of glossy black over-ear headphones. The camera begins tight on the headphones resting on a white table in a quiet, empty room — silence. A hand reaches in, picks them up, and puts them on. The moment they click onto the head, music kicks in, and the world transforms. The camera pulls back, revealing the person standing in a bustling coffee shop — a barista slides a cup across the counter in the foreground, steam rising, but the wearer walks past unfazed, eyes closed, to the music. The camera follows them seamlessly through a door that opens directly into a skatepark — a skater rolls past perfectly timed to the beat, the wearer walks between ramps without looking, and the camera tracks alongside them. They push through another door into a packed laundromat — machines spinning, someone folding clothes, a kid running past — all choreographed to the rhythm, all ignored by the wearer who is completely immersed. The camera swoops low under a hanging sheet and rises into a rooftop at golden hour — the city skyline wide open, wind hitting the wearer’s clothes. They stop at the edge, and the camera orbits them slowly for the first time. They open their eyes and smile. They take the headphones off — the music stops instantly, replaced by the rush of city noise, wind, and distant traffic. They look at the headphones in their hand. They put them back on. Music returns. They close their eyes again. The camera continues orbiting and slowly pulls upward into a wide aerial view — the wearer small on the rooftop, the city around them, everything moving except them. The headphones catch the last glint of golden light. Brand name appears in clean, minimal typography. The entire video is one continuous flowing shot — the camera never stops moving, never cuts, gliding through spaces connected by doors, walls, and seamless architectural transitions.”

Prompts para vídeos UGC

El contenido generado por el usuario funciona mejor cuando parece auténtico. Piensa en cámara en mano, entornos naturales y momentos cotidianos con los que cualquiera se identifica:

“15 seconds UGC style skincare review video, filmed on smartphone, natural bedroom window lighting, casual handheld selfie angle. A young woman with brown hair pulled back, natural skin with visible texture, wearing a casual grey t-shirt, in her cozy bedroom — books on shelves, plants on the windowsill, clothes on a chair, lived-in and real. She holds the @(img1) (LUNA Aurora Serum bottle) up to the camera. The video opens with her looking into the camera, excited expression: “Okay, so I’ve been using this for two weeks, and I need to talk about it.” Quick jump cut — she’s now showing the bottle closer to the lens, tilting it so the holographic text catches the light from the window: “The texture is insane, it’s like water but silky?” Jump cut — extreme close-up of her pressing the dropper, the serum dropping onto her fingertips, she rubs it between her fingers, showing the consistency. Jump cut — she’s applying it to her cheek in the mirror, phone propped up, you can see her reflection: “It absorbs in like two seconds, no stickiness.” Jump cut — she leans into the camera, pointing at her cheek with a genuine smile: “Look, I actually have a glow right now, and I’m literally wearing nothing.” Jump cut — she holds the bottle up one final time, gives it a little shake: «This is my new holy grail, I’m not even being dramatic.” She laughs, the video cuts. Throughout the video, the tone is genuine, unscripted-feeling, warm — she talks fast, uses natural pauses, laughs at herself. Each jump cut is slightly closer or at a different angle, as if she filmed multiple takes and edited the best bits together. The lighting is soft natural daylight, no ring light, no filters. The image is slightly imperfect — natural phone quality, not color graded, authentic. The sound is direct from the phone mic — room ambience, her natural voice, no music underneath. The overall feel is trustworthy, relatable, real — a friend telling you about something she genuinely likes.”

Prompts para videojuegos

Si tienes en mente un tráiler conceptual, una pieza cinematográfica o simplemente quieres ver cómo queda un personaje en movimiento, este tipo de prompt da muy buen resultado:

“15-second cinematic video game trailer, photorealistic, realistic rendering. Floating islands world: ancient energy cores failing, everything collapsing. Buildings crack, streets split, city blocks fall into a void. Protagonist: young, agile runner (island messenger), wears a lightweight armored suit with a magnetic grappling hook on the arm. Sequence: 1. Wide shot of a massive floating city breaking apart against a golden sky, architecture chunks tumbling in slow motion. 2. Runner sprinting across the rooftop as the building splits beneath their feet. 3. Leaping off a collapsing skyscraper edge, firing a grappling hook at passing debris, and swinging across a gap. 4. Rapid parkour cuts: wall-running on falling buildings, sliding under collapsing archways, vaulting over crumbling walls. 5. Mid-air combat against armored jetpack enemies over a glowing energy core, punches/kicks exchanged while free-falling between islands. 6. Slow-motion wide shot: runner falling through sky surrounded by drifting city debris (buildings, bridges, statues, trees) in freefall, beautiful and catastrophic. 7. Runner landing on new island, rolling, sprinting forward as cracks spread behind. 8. Final wide shot: dozens of collapsing floating islands against sunset sky, debris falling like rain. Cut to black. Title in fractured typography, letters slightly separating as if about to fall apart. Soundtrack: fast-paced orchestral mixed with electronic bass and percussion. Tone: epic, vertiginous, relentless. High production value.”

Escribir un buen prompt es, básicamente, dirigir sin cámara. Decides quién sale, qué hace, cómo se encuadra y qué se siente. Cuanto más claro lo tengas tú, más fiel será el resultado. Empieza con algo simple, prueba, cambia cosas y experimenta sin miedo. Ahí es donde aparece la magia.