How to write prompts for Seedance 2.0

If you’ve ever typed a prompt into an AI video generator tool and thought, “That’s… not what I meant,” you’re not alone. The gap between what you imagine and what the model delivers usually comes down to one thing: how you talk to it.

Seedance 2.0, ByteDance’s latest AI video generation model, helps close the gap. It interprets your words like a director reads a script. Camera angles, character actions, lighting, pacing, and even synchronized audio respond to how you structure your prompt. 

The model accepts text, images, video clips, and audio files as inputs. You can combine up to 14 assets in a single generation using its @tag reference system.

A powerful tool only works if you know how to guide it. Here’s how to write prompts that get the most out of Seedance 2.0.

Writing effective prompts for Seedance 2.0

The biggest shift with Seedance 2.0 is simple: you’re no longer just describing a scene, you’re directing one. That means your prompt needs to carry more intentional information than a simple visual description.

Here are the fundamentals. First, keep your prompts between 100 and 260 words. Shorter prompts tend to produce vague results. Longer ones can overwhelm the model and cause it to lose focus.

Second, be explicit about motion. The model can’t guess intensity from a still reference, so instead of writing “a car passes by,” write “a car speeds past from left to right, kicking up dust.” Degree adverbs like quickly, gently, dramatically, or frantically make a noticeable difference. 

Third, always include at least one style keyword like cinematic, documentary, commercial, anime, photorealistic. This anchors the overall visual direction. 

Finally, iterate one element at a time. If the action looks good but the framing feels off, adjust the camera description first. 

Seedance 2.0 standard prompt structure

After extensive testing across the community, one prompt framework has emerged as the most reliable for consistent results. It follows this order:

Subject + Action + Camera + Style + Constraints

• The subject defines who or what is in the scene. Be specific: age, clothing, expression, and posture all help the model build a clearer picture.
• The action describes what happens. Use the present tense and focus on one primary movement per shot. 
• The camera tells the model how the scene is framed (wide, medium, close-up) and how it moves (dolly-in, pan, orbit, handheld).
• The style sets the visual tone. This can include lighting, color grading, film references, or atmosphere. 
• Constraints reduce artifacts and maintain consistency. Examples include “no distortion,” “maintain face consistency,” or “steady motion.”

If you’re using references, Seedance 2.0 assigns labels like @Image1, @Video1, or @Audio1 to each upload. You can reference them directly in the prompt. For example: 

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

This improves consistency and prevents visual drift.

Using time stamps for precise timing

When your scene includes multiple beats, time stamps give you much more control over pacing and structure.

Instead of describing everything in one continuous block, break the sequence into timed segments. This helps the model understand exactly when a new action, camera move, or visual change should happen.

For example:

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

This approach works especially well for:

• Multi-shot sequences.
• Narrative build-ups.
• Action scenes with clear choreography.
• Style transitions within a single clip.

Time stamps reduce ambiguity. The model doesn’t have to guess what happens first or how long each moment lasts. You’re defining the rhythm of the scene, almost like editing on paper before generation.

Keep each segment focused on one main action. Clear timing leads to cleaner motion and smoother transitions.

Maintaining consistency with all-around references

Consistency is one of the main challenges in AI-generated video. If a key element shifts between shots, the whole sequence can feel unstable.

An all-around reference is a specific visual element that stays consistent throughout the entire video. This is usually something tangible and clearly defined, such as:

• A specific outfit or garment.
• A product with fixed design details.
• A distinctive accessory or object.
• A vehicle, weapon, or prop.
• A recurring architectural element.

This anchor should remain stable while other aspects of the scene evolve, such as camera movement, action, or environment.

When using reference assets, define their role clearly:

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

If your sequence includes multiple time-coded segments or camera changes, reinforce that the element must remain consistent throughout. This helps prevent visual drift and keeps the narrative cohesive.

You can also include short constraints like:

• “Maintain product design throughout.”
• “Keep outfit unchanged across shots.”
• “Preserve object proportions.”

Best prompts for Seedance 2.0

Strong prompts tend to be visually specific, action-driven, and technically clear. They tell the model what to show and how to show it. 

Below, you’ll find prompt examples organized by category. You can copy them, tweak the details, and adapt them to your project.

Cinematic narrative and drama prompts

Cinematic prompts benefit from emotional specificity and deliberate camera work. Think about how a scene would feel if you were sitting in a theater:

“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.”

Animation prompts

Seedance 2.0 handles stylized animation well when the visual direction is clearly defined.

“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.”

Fashion prompts Seedance 2.0

Fashion prompts work best when you emphasize fabric movement, lighting, and editorial composition, like in this example:

“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.”

Motion prompts

If your main goal is dynamic movement, lean into strong action verbs and pacing cues. This works well for athletic, mechanical, and natural motion.

“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.”

Music video prompts

This type of prompt should reference rhythm, editing pace, and visual energy. If you have an audio file, use the @Audio tag to sync movement with the 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.”

Seedance 2.0 commercial and product prompts

Product-focused prompts demand clean composition, controlled lighting, and smooth mechanical motion.

“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.”

UGC video prompts

User-generated content works best when it feels real. Think handheld cameras, natural settings, and relatable moments.

“User-generated content works best when it feels real. Think handheld cameras, natural settings, and relatable moments.“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.”

Videogame prompts

Game-related prompts work great for concept trailers, cutscene prototypes, and character showcases. Try these prompt examples:

“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.”

Prompting Seedance 2.0 is less about thinking like a filmmaker. Define your subject, choreograph the action, choose your camera, and set the mood. The more intentional your instructions, the closer the output will match your vision. Start simple, iterate fast, and don’t be afraid to experiment. That’s where the best results come from.