Seedance 2.0: prompt patterns and six use cases for AI video

Seedance 2.0 is the model creators reach for when the brief reads like a director's brief. It accepts text, up to nine images, three reference videos, and three audio files in one request, addressable with @image1, @video1, and @audio1 syntax. This guide covers the prompt grammar that respects the model, its current OmniArt limits, and six tested use cases with prompts and results.

What Seedance 2.0 is

Seedance 2.0 generates 4–15-second clips with native audio. Standard reaches 4K; Fast and Mini reach 720p. The headline is still the multi-reference architecture: one request can combine images, videos, and audio, provided it contains at least one visual reference.

Why the multi-reference system matters

Most video models accept one reference, or none. Seedance 2.0 accepts a stack and lets you bind each reference to a role inside the prompt. Use @image1 for the character's face, @image2 for the costume, @image3 for the location, @video1 for the camera move you want, @audio1 for the music bed. The output respects each as a discrete instruction instead of averaging them into noise.

That's the practical reason character likeness holds across shots: the same @image reference goes into every shot in the timeline, and the model uses it as the identity anchor rather than re-inferring the character from the prompt each time.

Prompt structure that works

Seedance 2.0 rewards a six-part structure.

1. Subject — who or what is on screen
2. Action / movement — what they do
3. Setting / environment — where it happens
4. Visual style — film references, palette, era
5. Camera direction — specific cinematography terms
6. Lighting — direction, quality, time of day

A good template prompt:

"Subject (with @image1 reference if applicable). Action. Setting. Visual style. Camera direction (specific cinematography term). Lighting detail."

Multi-shot timeline notation

Seedance 2.0 responds to shot order, not second-level timestamps. Use numbered shots and describe each beat in sequence.

Shot 1: wide establishing shot, character (in @image1) walks into the scene
Shot 2: medium tracking shot follows them across the room
Shot 3: 360-degree orbit around the table they reach

Pin the same @image1 across every segment. Likeness stays consistent through the cut.

Reference tagging discipline

A short rulebook that pays off:

• Use @image1, @image2 for face photos and product shots.
• Use @video1 for the camera move you want copied.
• Use @audio1 when the audio bed matters more than the model's default.
• Reference each tag explicitly in the text. Don't rely on the model to infer which reference is which role.

Six tested use cases with prompts

Each prompt below is one we've run on Seedance 2.0. The results column is what we got, with generation time measured on Standard 720p.

1. Cinematic film scene

"A retired detective in a long dark coat walks through a rain-soaked alley at night. Neon signs reflect red and blue on the wet cobblestones. He pauses, lights a cigarette, and glances over his shoulder. Slow push-in from wide shot to medium close-up. Film noir style, anamorphic lens flare, teal-orange color grading, film grain."

Result. Smooth camera push-in. Convincing rain reflections, natural coat movement. Cigarette lighting renders without hand distortion. Rain and city ambient audio generated in sync. ~70 seconds.

2. Product commercial

"A luxury perfume bottle rotates slowly on a black marble surface. Golden liquid catches the light as it turns. Soft particles of gold dust float in the air around it. Macro close-up, slow 360-degree orbit camera. Studio lighting with warm rim light, high-end commercial photography style."

Result. Glass refraction and liquid behavior accurate. Particle drift natural. Smooth full rotation, correct light angles, marble texture visible. ~65 seconds.

3. Music video

"A female singer in a flowing red silk dress performs on a rooftop at sunset. City skyline stretches behind her. Wind blows her hair and dress dramatically. She sings with emotional intensity, arms spread wide. Dynamic tracking shot circling around her. Golden hour backlighting, lens flare, vibrant warm tones."

Result. Realistic dress physics. Fluid tracking orbit. Face stays consistent through the rotation. Hair movement matches wind direction. Generated ambient musical track. ~75 seconds.

4. Character portrait in motion

"An elderly Japanese craftsman in a traditional wooden workshop, morning light streaming through paper screens. He slowly lifts a hand-forged ceramic tea bowl, examining it with quiet pride. His weathered hands rotate the bowl gently. Close-up of his hands, then slow tilt up to reveal his face. Wabi-sabi aesthetic, warm natural light, documentary portrait quality."

Result. Correct finger count. Natural joint movement. Smooth tilt from hands to face. Realistic light through screens. Faint workshop ambient sounds. Realistic skin texture. ~80 seconds.

5. Nature and landscape

"Aerial drone shot gliding over a misty mountain valley at sunrise. Layers of fog roll between emerald green peaks. A winding river reflects the golden morning light below. Eagles soar through the frame at eye level. Smooth forward tracking with slight descent. Epic landscape, volumetric fog, golden hour lighting."

Result. Independent fog layers create convincing depth. River reflections update with camera position. Strong palette balance. Volumetric fog renders cleanly. Wind and bird call audio. ~55 seconds — the fastest of the six.

6. Anime and fantasy

"An anime warrior princess stands atop a cliff overlooking a burning medieval city at night. Her long silver hair and crimson cape billow in the wind. She draws a glowing blue katana, electricity crackling along the blade. Cherry blossom petals swirl around her. Dynamic low-angle shot with slow push-in. Cel-shading style, vibrant neon accents, dramatic speed lines."

Result. Consistent cel-shading throughout. Fluid katana draw. Electricity effect integrates naturally. Independently moving cherry blossoms. Firelight interaction with cape. Dramatic sword swoosh audio. ~70 seconds.

Common errors and fixes

Seedance 2.0 vs Seedance 1.0

If you've used 1.0, the gap to 2.0 is wider than the version number suggests.

When to choose something else

Seedance 2.0 isn't the right tool for every brief.

Pricing on OmniArt

Seedance 2.0 is priced per output second inside the OmniArt video workspace. At 720p, Standard costs 14 credits per second, Fast costs 11, and Mini costs 8. A five-second clip therefore costs 70, 55, or 40 credits respectively. Standard requires Creator or Ultra; Fast and Mini require Starter or higher. Ultra currently includes no-credit asset generation rather than a percentage discount.

Getting started on OmniArt

Seedance 2.0 Standard, Fast, and Mini sit inside the OmniArt video workspace alongside MiniMax H3, Gemini Omni, Happy Horse, Kling, Grok Imagine, Veo 3.1, Sora 2, V6, and other current video models. They share the same balance and creation history, while each model keeps its own supported parameters and reference limits.

Start with the cinematic film scene prompt above to feel out the multi-reference workflow, then move to the music-video brief once you want to test face consistency across motion.

If you're choosing between Seedance 2.0 and HappyHorse 1.0, the HappyHorse 1 vs Seedance 2 comparison walks through the trade-offs shot by shot. For the next-generation workflow, continue with what changed in Seedance 2.5 and select the task-specific prompting guide from there.