Seedance 2.0 Prompt Guide: Formulas, Examples & Multimodal Tips

Seedance 2.0 is powerful—but weak prompts still produce weak clips. The difference between a random AI video and a directed one almost always comes down to how clearly you describe what each input does and what should happen on screen.

This guide is a practical rewrite focused on prompting: a reusable formula, complete input → prompt → output examples for every mode, and the multimodal @ reference patterns that unlock Seedance 2.0’s real strength.

If you also need a product overview and step-by-step UI walkthrough, see the Seedance 2.0 complete guide and how to use Seedance 2.0.

What Seedance 2.0 Expects From a Prompt

Seedance 2.0 (by ByteDance) is a multimodal AI video model. It does not only read a sentence—it can combine:

• Text — story, action, camera, mood
• Images — character look, product, style, storyboard frames
• Video — motion, camera language, timing, transitions
• Audio — music pacing, SFX mood, or voice tone reference

Your job as the prompt writer is to assign roles. Tell the model which asset controls appearance, which controls motion, and which controls rhythm. When those roles are clear, results become consistent.

Seedance 2.0 Input & Output Specs (Know These First)

Before writing prompts, lock in what the model can actually accept and return on Van Gogh Studio:

Rule of thumb: more references are not automatically better. Prefer 1–4 strong anchors over a crowded upload stack.

The Core Prompt Formula

Use this structure for almost every Seedance 2.0 generation:

[Subject] + [Action / Timeline] + [Environment] + [Camera] + [Style / Lighting / Mood] + [@Asset roles]

Mini formula by mode

• Text-to-video: build the whole scene in language (no @ needed).
• Image-to-video: describe motion + camera; do not re-describe the whole image.
• Video reference / video-to-video: split into Preserve vs Change.
• Multimodal: declare a hierarchy (@Image1 look, @Video1 motion, @Audio1 pacing).

Prompting Best Practices

1. Lead with the subject. Put the main character or product first.
2. Write visible motion. Prefer verbs the camera can show: turns, walks, lifts, drifts, tracks.
3. Use a timeline for multi-beat clips. Segment as 0–3s / 4–8s / 9–13s when the story shifts.
4. Separate reference vs edit. “Reference the camera of @Video1” ≠ “Replace the person in @Video1.”
5. One primary intent per sentence. Avoid packing five conflicting styles into one line.
6. Say what you want, not what you hate. Skip vague negatives like “no weird hands”; rewrite as positive constraints (“clean hands resting naturally on the table”).
7. Iterate one variable at a time. Change camera or costume or pacing—not all three at once.

Mode 1: Text-to-Video Prompts

Best for: concepts from scratch, mood tests, story beats with no reference assets.

Input → Prompt → Output

Stronger text-to-video pattern (timeline)

0–4s: Wide shot of an empty desert highway at blue hour. Heat haze shimmers above the asphalt.
5–10s: A vintage red convertible drives toward camera, dust trailing behind the tires.
11–15s: Camera rises into a gentle crane-up as the car passes underneath; warm headlights streak across frame.
Style: photoreal, anamorphic bokeh, contemplative, filmic color grade.

Text-to-video tips

• Name shot size (wide / medium / close-up) and camera move (pan, track, orbit, push-in).
• Specify lighting (soft dawn, harsh neon backlight, candle flicker).
• Describe pacing (slow meditative vs fast high-energy).
• Keep one primary subject unless you intentionally stage a two-shot.

Mode 2: Image-to-Video Prompts

Best for: animating a photo, concept art, product still, or character sheet.

Here the image already defines appearance. Your prompt should choreograph motion layers, not rebuild the scene from zero.

Example A — Atmospheric scene animation

Example B — Portrait micro-motion

Example C — Style-preserving art animation

Image-to-video motion layers checklist

• Foreground: mist, leaves, candle flame, water ripples
• Midground: subject action (head turn, hand raise, walk cycle)
• Background: clouds, flags, distant crowd, light flicker
• Camera: locked-off / slow push-in / subtle drift / gentle handheld

High-impact image-to-video vocabulary

Mode 3: Video Reference / Video-to-Video Prompts

Best for: keeping motion/timing from a source clip while changing style, wardrobe, or world.

Always answer two questions before writing:

1. What must stay? (motion, framing, timing, dialogue beats)
2. What must change? (costume, location, color grade, genre)

Preserve vs transform template

Preserve: Keep the dancer’s choreography, timing, body posture, and original camera framing from @Video1 exactly.
Transform: Restyle the environment into an ethereal forest glade. Costume becomes a translucent flowing gown. Soft teal and lavender tones, volumetric god rays, floating fireflies, light film grain.

Style transfer example (film noir)

Preserve the subject’s walk, pace, and tracking camera from @Video1 entirely.
Transform into a 1940s film-noir detective scene: trench coat and fedora, rain-slick cobblestones, glowing gas lamps, high-contrast black-and-white grade, fog, slight 35mm flicker. Mysterious, brooding tone.

Video continuation / extension example

Continue seamlessly from the final frame of @Video1.
She steps through the doorway; expression shifts from curiosity to wonder.
Reveal a vast library of impossible scale—shelves stretching upward, glowing manuscripts.
Warm golden light. Slow reverent steps; camera tilts up with her gaze.

Mode 4: Multimodal Prompts (@Image + @Video + @Audio)

Best for: production control—character lock + motion copy + music-driven pacing.

This is Seedance 2.0’s core workflow. Upload assets, then assign each one a job with @.

How @ referencing works

In the prompt, call assets by name and role, for example:

• @Image1 = character appearance / first frame
• @Image2 = costume or product detail
• @Video1 = camera language or action timing only
• @Audio1 = pacing / music energy

If you do not assign roles, the model may blend assets incorrectly.

Example A — Hierarchy (one visual authority)

Example B — Equal fusion of two looks

Example C — Audio as the director

Multimodal prompt checklist

• Every uploaded asset that matters is mentioned with @
• Each asset has one clear job (look / motion / style / pacing)
• Conflicts are resolved (e.g. “ignore @Video1 colors”)
• One unifying aesthetic ties the mix together
• Audio (if used) is linked to camera energy or cut rhythm

Copy-Paste Prompt Templates

1) Product hero shot (image + text)

Use @Image1 as the product hero. Keep logo, shape, and label text stable.
0–3s: Slow orbit around the product on a clean reflective surface.
4–8s: Soft light sweep reveals material texture.
9–12s: Gentle push-in to the label details.
Style: premium commercial, crisp reflections, controlled studio lighting.

2) Character consistency across a short scene

@Image1 defines the character’s face, hair, and outfit—do not alter identity.
0–4s: Medium shot, character looks toward camera, subtle smile.
5–10s: Character turns and walks into the rainy street; camera tracks beside them.
Lighting: cool neon rim light + warm shop windows. Cinematic, coherent, no face morphing.

3) Camera copy from reference video

@Image1 is the first frame / subject look.
Fully reference the camera movement and transition rhythm from @Video1 only.
Do not copy costumes or background from @Video1.
Scene: dusk rooftop city view, wind in coat, distant traffic lights bokeh.

4) Storyboard-to-sequence

Based on the storyboard frames in @Image1–@Image4, generate a continuous 12-second opening.
Follow shot order, camera angles, and scene progression shown in the boards.
Tone: warm nostalgia, soft daylight, gentle dissolves between beats.
Keep character design consistent with @Image1.

Common Prompt Mistakes (and Fixes)

How to Run These Prompts on Van Gogh Studio

1. Open the AI video studio and select Seedance 2.0 (or open the Seedance 2.0 model page).
2. Choose text / image for simple jobs, or reference / multimodal when using multiple images, video, or audio.
3. Upload references, then write your prompt with clear @ roles.
4. Set duration (4–15s), aspect ratio, and resolution.
5. Click Create, review the clip, then iterate one variable at a time.

For the full UI walkthrough, use How to use Seedance 2.0.

Quick FAQ

Do I need @ tags for text-only generation?

No. Use @ only when you uploaded images, video, or audio and need to assign roles.

Should image-to-video prompts describe clothing and face again?

Usually no. The image already carries appearance. Focus on action, camera, atmosphere, and style preservation.

How long should a Seedance 2.0 prompt be?

Long enough to cover subject, action, camera, and mood—often 2–6 sentences, or a short timeline. Clarity beats length.

Can I control multi-shot storytelling in one clip?

Yes. Use a timed structure (0–3s / 4–8s / …) and, when available, storyboard images with explicit shot order.

What’s the fastest way to improve weak outputs?

Tighten the subject + action first, then add one camera instruction, then add references with explicit roles. Avoid stacking new ideas every retry.

Conclusion

Great Seedance 2.0 prompts are director’s notes, not poetry. Specify the subject, stage the action on a timeline, choose a camera plan, and—when using multimodal inputs—assign every asset a job with @.

Start with the formula in this guide, copy a template close to your use case, and iterate one variable at a time. For deeper feature context, read the complete Seedance 2.0 guide, then generate on Van Gogh Studio.