Seedance 2 — ByteDance's flagship video model — rewards structured prompts the way a cinematographer rewards a clear shot list. Vague prompts produce generic output; specific prompts produce the kinetic, confidently-composed clips the model is actually capable of. This guide is the practical version: a six-part prompt anatomy, the camera and motion phrases Seedance 2 parses cleanly, style language, and 20 copy-paste templates you can drop into the Oakgen AI Video Generator today.
What Seedance 2 does best
Seedance 2 has distinct strengths — prompts that lean into them get noticeably better results.
Kinetic motion. Chases, skateboarding, flowing fabric, crashing waves — anything with speed and directional energy. The model has a strong sense of momentum and keeps subjects consistent while the environment moves around them.
Action and physics. Collisions, pouring liquids, falling objects, explosions, cloth, hair in wind. The physics isn't perfect, but it is confident — confident physics reads better than tentative physics.
Fast generation. Seedance 2 is one of the faster tier-1 models on Oakgen. When you can iterate five prompts in the time Veo takes one, you end up with better shots simply because you tested more directions.
Prompt adherence for composed scenes. Describe the frame carefully — subject placement, foreground-background, angle — and Seedance 2 largely delivers it. Weak spots: facial expression, lip sync, multi-hand interaction (covered at the end).
The prompt anatomy
Every Seedance 2 prompt that consistently works follows the same six-part structure.
1. Subject. Who or what is in the frame. Not "a person" but "a woman in her thirties with cropped dark hair, wearing a charcoal wool coat."
2. Action. Active present-tense verb. Not "standing there" but "turning her head slowly toward the window."
3. Environment. Lighting, time of day, location. "A rain-soaked Tokyo alley at 2am, neon signs bleeding color onto wet asphalt."
4. Camera. How the shot is framed and how it moves — where most prompts fall apart. Covered below.
5. Motion. The speed and character of everything moving. Steady? Frantic? Slow motion? Time-lapsed?
6. Style. The film look. Anamorphic? 35mm? A cinematographer or director as shorthand.
A worked example with all six:
A surfer in a black wetsuit [subject] paddling hard into a forming wave [action],
on a foggy gray morning at a Pacific Northwest beach, dawn light, soft diffused
sky [environment], low-angle handheld tracking shot from the water, just above
sea level [camera], real-time speed building to slow motion as the wave
crests [motion], shot on anamorphic 35mm, Roger Deakins cinematography,
muted color grade [style].
That clip will land. "Surfer catching a wave, cinematic" will not.
Camera directives Seedance understands
Seedance 2 responds cleanly to specific camera terminology. Use the actual film-industry phrases rather than plain English.
| Directive | Result |
| --- | --- |
| dolly in / dolly out | Smooth forward/backward push on subject |
| truck left / truck right | Lateral slide across scene |
| crane up / crane down | Vertical rise/descent with slight arc |
| aerial shot / drone shot | High overhead with forward drift |
| handheld tracking | Organic, slightly shaky follow |
| locked-off wide | Static wide frame, no movement |
| whip pan | Fast horizontal pan with motion blur |
| orbit shot | Camera circles the subject |
| push in on the eyes | Slow micro-zoom on a face |
| dutch angle | Tilted frame for tension |
| over-the-shoulder | Classic OTS framing |
| POV | First-person viewpoint |
Stack two at most — "handheld tracking, low-angle" works. Four confuses the model.
Seedance 2 generates 5–15 second clips. To stitch them into scenes with consistent shot language and characters, use Cinema Studio on Oakgen — it handles multi-shot continuity, storyboarding, and scene-level prompt reuse so you are not re-describing your character in every clip.
Motion language
The biggest upgrade most people can make is describing motion with specificity. The model has a strong sense of speed — but only if you give it one.
Bad: "leaves falling" Good: "slow-motion falling leaves at 120fps, each leaf rotating lazily against a blurred forest background"
Bad: "guy running through the woods" Good: "fast handheld chase shot following a man sprinting through a dense forest, branches whipping past the lens"
Bad: "clouds in the sky" Good: "rolling storm clouds compressed to 5 seconds of time-lapse, shadows racing across a wheat field below"
Patterns worth stealing:
- Speed anchors: "real-time," "slow motion at 120fps," "time-lapse compressed to 4 seconds," "frozen except for…"
- Secondary motion: describe what the environment is doing too — hair, dust, fabric, water, shadow. Seedance handles secondary motion well and it massively improves realism.
- Motion verbs, not adjectives: "the fabric snaps in the wind" beats "windy fabric."
Style phrases
Style is where prompts go from "clearly AI" to "this could be a real short." Seedance 2 recognizes cinematographers, directors, film stocks, and lens types.
- Film stock:
shot on 35mm film,super 16mm,Kodak Portra 400 film look,Fuji Eterna color palette - Lens language:
anamorphic lens, subtle horizontal flares,wide 14mm with barrel distortion,85mm portrait lens, shallow depth of field - Cinematographer shorthand:
Roger Deakins cinematography(crisp, naturalistic),Emmanuel Lubezki(natural light),Bradford Young(low-light warmth) - Director looks:
Wes Anderson symmetrical composition, pastel palette,Denis Villeneuve scale and negative space,David Fincher cool desaturated palette - Color grade:
teal and orange,bleach bypass,muted earth tones,high-key pastel
One film reference plus one technical spec is the sweet spot. Three or more cancel each other out.
20 copy-paste templates
Drop any of these into the Oakgen AI Video Generator, swap the bracketed slots, and generate.
Product videos
Unboxing hero. Commercial-style reveal.
Extreme close-up of two hands slowly lifting the lid off a matte black [PRODUCT]
box on polished concrete, soft top-down key light, slow macro push-in at 60fps,
shallow depth of field, 50mm macro lens, Apple-style commercial, clean grade.
Tip: Keep backgrounds neutral — busy environments fight the product.
Hero product shot. Static elegance with subtle motion.
A single [PRODUCT] rotating slowly on a seamless charcoal gradient backdrop,
dramatic rim light back-right, soft key front-left, locked-off medium close-up,
anamorphic 35mm, subtle dust particles drifting in the beam.
Tip: "Subtle dust particles" is the cheat code — adds life without distracting.
Product in use. Lifestyle and demo content.
Over-the-shoulder shot of a woman in her early thirties holding a [PRODUCT]
in a sunlit modern kitchen, morning golden hour through sheer curtains,
handheld 35mm feel, real-time, Fuji Eterna stock, shallow focus, soft bokeh.
Tip: Specify the user — vague users read as stock footage.
Lifestyle scene. Aspirational social.
A [PRODUCT] on a weathered wooden table next to an open notebook and a
steaming ceramic cup, afternoon light through a rain-streaked window,
locked-off wide, slow 5-second push in, 35mm Kodak Portra 400, warm amber grade.
Tip: Environmental storytelling beats product-first framing on social.
Social media
Reel opener. Instagram/TikTok thumb-stop.
Fast whip pan from a blurred neon cityscape into a crisp medium shot of a
young creator in a bold oversized jacket, direct-to-camera eye contact, quick
smile, urban night with bokeh string lights, anamorphic lens, high-contrast grade.
Tip: Whip pans at the start of a Reel measurably lift completion rate.
TikTok kinetic. Clip-to-clip energy.
Handheld tracking running alongside a skateboarder carving through a sunlit
downtown plaza, low-angle, golden hour, motion blur on the wheels, real-time
with brief slow-motion on the trick landing, 16mm film aesthetic, vibrant grain.
Tip: Mix real-time with a beat of slow motion — reads as intentional cutting.
YouTube intro. Cinematic channel open.
Slow crane-down reveal from a foggy mountain ridge to a lone hiker in a red
jacket stepping into frame, dawn blue hour, atmospheric mist, anamorphic 35mm,
Roger Deakins cinematography, muted earth tones.
Tip: Reveal the subject in the final third of the clip, not frame one.
Story slide. Vertical 9:16 brand content.
Vertical 9:16 frame, medium close-up of a hand pouring pour-over coffee into
a white ceramic cup in slow motion at 120fps, steam rising and swirling,
morning window light from the left, 85mm, shallow depth of field, warm grade.
Tip: Seedance handles liquid well — lean into it for vertical content.
Cinematic
Trailer opener. First-shot-in-a-trailer feel.
Locked-off extreme wide of a lone figure at the edge of a vast salt flat at
dawn, silhouetted against a pastel sky, almost no motion except wind rippling
their coat, 65mm large format, Denis Villeneuve scale, slow 8-second hold.
Tip: Extreme wide + near-stillness + slow duration = cinematic authority.
Establishing shot. For setting a scene.
Slow aerial drone gliding forward over a dense pine forest in the Pacific
Northwest at foggy dawn, a single winding road cutting through, overcast
diffused light, real-time, cinema drone, muted green and gray, Fincher grade.
Tip: Drone shots are the most reliable cinematic mode in Seedance 2.
Emotional beat. Character moment.
Tight close-up on a woman's face near a rain-streaked window, warm interior
lamp light from camera-left, eyes welling without crying, quiet realization,
locked shot with micro push-in, 85mm portrait, teal-and-orange, 35mm grain.
Tip: Keep prompt focus on one micro-expression — never two.
Character intro. Establishing a protagonist.
Slow handheld tracking following a figure in a long black coat walking away
from camera down a foggy cobblestone street at night, warm sodium streetlamps,
breath visible, anamorphic lens with subtle flares, desaturated noir grade.
Tip: Walking away is easier than walking toward — use it for the intro, save the reveal for shot two.
Explainer
Process visualization. Tutorials and demos.
Locked-off overhead top-down of hands assembling [OBJECT] on a clean white
surface, even soft lighting, real-time precise movement, macro lens, clinical
commercial aesthetic, clean high-key color grade.
Tip: Top-down plus high-key lighting is the explainer default.
Data animation. Abstract concept visuals.
Abstract flowing data particles forming a glowing network graph in 3D space
against dark navy, particles moving along curved paths with subtle bloom,
slow dolly through the structure, neon cyan and magenta, motion graphics style.
Tip: For abstract work, name the palette — Seedance defaults to muddy if you don't.
Timeline visualization. Historical/sequential content.
Time-lapse of a city street corner transitioning from 1950s to present day
over 8 seconds, locked-off wide, passing cars and pedestrians blurring through
eras, color grade shifting from sepia to modern neutral.
Tip: "Transitioning over [X] seconds" gives the model a concrete pacing target.
Metaphor visual. Opening a concept-driven video.
A single glass marble dropping into a still dark pool in extreme slow motion
at 240fps, ripples expanding in perfect concentric circles, hard rim light
from behind, locked-off macro, deep navy water, 100mm macro, minimal grade.
Tip: Physics metaphors (drops, impacts, ripples) are Seedance's strongest abstract mode.
B-roll
Nature atmospheric. Establishing outdoor mood.
Slow aerial drift over a misty redwood forest at dawn, golden shafts of light
cutting through fog between massive trunks, slow forward movement, cinema drone,
muted forest green and amber, National Geographic aesthetic.
Tip: "Shafts of light" is a reliably strong visual cue in Seedance.
City establishing. Urban B-roll.
Locked-off wide of a busy Tokyo intersection at dusk, neon signs reflecting
on wet pavement after rain, pedestrians with umbrellas, time-lapse compressed
to 6 seconds, anamorphic 35mm, saturated teal-and-magenta cyberpunk grade.
Tip: Time-lapse plus wet pavement is a reliable Seedance win for urban shots.
Close-up texture. Detail inserts and cutaways.
Extreme macro close-up of raindrops hitting and rolling down dark green
leaves, slow motion at 240fps, soft overcast light, subtle motion in breeze,
100mm macro, shallow depth of field, rich saturated grade.
Tip: Macro + slow motion is the single highest hit-rate category in this guide.
Atmospheric mood. Transitions and interludes.
Slow push through a dust-filled abandoned warehouse, shafts of light through
broken windows, particles drifting in the beams, no subject in frame, slow
dolly forward, anamorphic lens, desaturated amber-and-gray, Lubezki aesthetic.
Tip: Subjectless shots are underused — perfect connective tissue between scenes.
If a Seedance 2 prompt hits, save the exact string — don't trust memory. The Oakgen AI Video Generator keeps your generation history, and Cinema Studio lets you reuse saved prompts across multi-shot scenes so your entire short has consistent language.
Where Seedance 2 struggles
Honest limitations — knowing these saves iterations and credits.
Fine facial expression. Subtle emotional shifts, especially in non-close-up shots, often flatten. Veo 3 is noticeably stronger for character fidelity and emotion.
Dialog lip sync. Seedance 2 syncs audio but is not best-in-class for dialog. If lip sync matters, reach for Kling 3.
Complex hand interactions. Multi-finger object manipulation is the weak spot shared by every current model. Cut away, crop out, or keep hands simple.
Precise motion control. If you need exact trajectory control — "step left then pivot right on beat 3" — Kling 3 with motion brush is currently better.
For most creators: Seedance 2 for kinetic, environmental, and product shots; Kling for motion-precise character work; Veo for expressive close-ups. See the Veo vs Kling vs Wan comparison for model selection and the Seedance 2 complete guide for deeper specs.
FAQ
How long should a Seedance 2 prompt be? Between 40 and 90 words is the sweet spot. Hit every element of the six-part anatomy and stop.
Does Seedance 2 support negative prompts? Not in the traditional sense. Instead of "no blurry faces," describe what you do want — "sharp focus on the subject's face." The model listens to positive description far better than exclusion.
Why does my prompt produce different results each time? All video models are stochastic. Run 3–4 generations and pick the best take — Seedance 2 is fast enough that this is cheap. See Oakgen pricing for credit costs.
Can I use a reference image? Yes. Image-to-video is supported, and the prompt structure above still applies — treat the image as a pinned subject and let your prompt drive camera, motion, and style.
Is there a referral program? Yes — the Oakgen referral program pays out on signups and subscriptions. Creators who publish prompt guides and tutorials tend to do well with it.
Seedance 2 is one of the most capable video models available right now, but capability only shows up when the prompt gives it something to work with. Open the Oakgen AI Video Generator, paste one of the templates in, and start iterating.
