Seedance 2.0 Prompt Writing Tips

Public tutorials and the official blog suggest that clear structure and specific wording improve results. Below we summarize commonly recommended formulas and the @ reference system; your platform may implement these in slightly different ways.

Last updated: 2026-03-27

Last verified: 2026-03-27

Refresh cadence: Every few days

Source basis and reading boundary

These guides are written as third-party reference summaries, not official product documentation or support content.

Sources used

Re-checked against the current ByteDance Seedance 2.0 project page, the Seed Models page, Dreamina help/resources, and BytePlus / ModelArk docs on March 24, 2026.

Boundary

Use these pages to understand public claims, common workflows, and terminology. Do not read them as official support, authorization, or product-owner statements.

Timeliness

Access routes, input limits, queue behavior, pricing, and API availability can change by surface. Treat Dreamina, BytePlus / ModelArk, and partner routes as separate products until current docs confirm otherwise.

Source basis

This page summarizes publicly available materials. Specs, pricing, and access may change, so verify with primary sources before making decisions.

ByteDance Seedance 2.0 project page
official · 2026-03-27
Dreamina official guide: How to use Seedance 2.0
platform · 2026-03-27

Prompt templates

Prompt template cluster

Use the dedicated prompt-template cluster for reusable templates, daily Input / Output updates, and future media evidence.

Coming soon (no assets yet)

Explore prompt templates

Golden Prompt Formula

The recommended Seedance 2.0 prompt formula: @asset reference + role assignment + scene description + camera/motion + style/lighting + sound design (optional) + timeline (optional). Example: '@Image1 as the character, walking through a neon-lit Tokyo alley at night, slow dolly forward, cinematic color grading, warm highlights, ambient rain sounds, 0-3s establish scene then 3-8s character turns to camera.' This seven-element structure, widely cited in public guides, maximizes control over the output. Start with the first three elements and add more as needed.

Prompt structure (formula)

Many guides recommend a seven-part structure: subject + action + environment + camera movement + lighting/mood + style + (optional) sound. Shorter variant: subject + action + scene + camera language + style + quality + constraints. Example: “A woman in a red coat walks through a snowy street, light snow falling, camera tracking beside her, cinematic lighting, 2K, sharp details.” Avoid vague words like “beautiful” or “cool”; use concrete details (colors, positions, textures).

@ reference system (multimodal)

When you upload images, videos, or audio, Seedance 2.0 supports referring to them in the prompt with @ tags (e.g. @Image1, @Video1, @Audio1). Examples from public docs: “@Image1 as the first frame,” “Reference @Video1 for camera movement,” “Use @Audio1 for background music.” This lets you control which asset drives composition, motion, or sound.

Motion and camera

Describe both subject motion and camera motion. Recommended terms include: slow zoom in (dolly in), pull back, truck (lateral move), orbit, tracking shot, static camera, handheld. Keep camera moves simple and stable for more reliable output. For action, words like “slow, smooth, continuous, natural” often yield better consistency; complex multi-person interaction or extreme motion can be harder to control.

Multi-shot temporal cues

For multi-shot sequences, public guides suggest using temporal language to structure the narrative: e.g. "First... Then... Finally..." or "Scene 1: ... Scene 2: ...". This helps the model plan transitions and maintain logical flow. Combine with consistent character references (e.g. "same character as @Image1") for best results.

What to avoid

Common failure causes in public guides: overly vague prompts; missing motion or camera; contradictory instructions (e.g. close-up and wide shot in one shot); no mood or lighting; no sound design when you want audio. For character consistency, add constraints like “same character, consistent outfit, natural anatomy, stable face.”

Advanced template board (coming soon)

This page now hosts the advanced prompt-template board. We keep Input/Output placeholders live in all locales and will add daily verified image/video evidence once materials are produced.

Examples & sources

Seven-part prompt example

Per public guides, the seven-part structure (subject + action + environment + camera + style + sound) is often recommended. Below is a typical example from third-party tutorials.

A woman in a red coat walks through a snowy street, light snow falling, camera tracking beside her, cinematic lighting, 2K, sharp details.

Dreamina experience

@ reference multimodal example

After uploading images or video, refer to them in the prompt with @ tags. Per public docs, these are common patterns.

@Image1 as the first frame; Reference @Video1 for camera movement; Use @Audio1 for background music.

ByteDance Seedance 2.0 project page

Advanced prompt board placeholder

Coming soon: daily validated prompt runs for "character", "scene", and "display/synthesis" categories with evidence links.

Input template -> Run -> Analyze -> Categorize -> Publish (image/video proof coming soon).

Frequently asked questions

How long should my prompt be?

One to three clear sentences is a good range. Include subject, action, camera, and style; add detail only where it affects the result. Too short is vague; too long can dilute the main idea.

Why does my output look wrong?

Common causes: vague prompts, missing motion or camera, or contradictory instructions. Add concrete details (colors, positions, camera moves) and avoid conflicting terms. See our best practices guide for more.

Seedance 2.0 Best Practices — Pro Tips for Better Video Output

How do I reference uploaded images?

Use @ tags (e.g. @Image1, @Video1, @Audio1) in your prompt. Examples: '@Image1 as the first frame' or 'same character as @Image1'. See our multimodal guide for the full @ system.

Seedance 2.0 Omni-Reference & Multimodal Input — Images, Video & Audio References Explained

What camera terms work best?

Public guides recommend: dolly in, pull back, truck, orbit, tracking shot, static camera, handheld. Keep moves simple and stable for more reliable output.

What prompt structure works best for multi-shot consistency?

Public guides recommend: subject + action + environment + camera + style, with explicit 'same character as @Image1' in every shot. Use temporal cues like 'First... Then... Finally...' and keep 1–2 reference images per character. See our best practices guide for more.

Seedance 2.0 Best Practices — Pro Tips for Better Video Output

Related capabilities

Related guides

Explore more guides