Official Guide

Seedance 2 Multi-Modal Video Guide

Seedance 2 supports four input modalities — image, video, audio, and text. Learn how to combine them with @ references for precise creative control.

Open Seedance 2 →

Quick Start

Two Creation Modes

🖼️ First / Last Frame

Upload a single image as the first frame (or last frame) plus a text prompt. Best for simple image-to-video animations where you have a clear starting point.

🌀 Universal Reference

Upload any combination of images, videos, and audio files, then use @ syntax to assign roles to each asset. The most powerful mode for multi-modal creation.

Upload Your Assets

Upload images (up to 9), videos (up to 3, total ≤15s), and audio files (up to 3 MP3s, total ≤15s). The combined file limit is 12 per generation. Prioritize assets that have the most impact on your desired result.

Write Your Prompt with @ References

Use the @ syntax to tell Seedance 2 how to use each asset. For example:@image1 as first frame, @video1 for camera movement, @audio1 for background musicType @ in the input field to see a list of available references.

Set Parameters

Choose your generation duration (4–15 seconds), aspect ratio, and quality mode. Note: if you're extending a video, the duration setting applies to the newly generated portion, not the total length.

Generate & Download

Click generate and wait 1–3 minutes. Seedance 2 will process all your multi-modal inputs and deliver a cinematic video with built-in audio. Download instantly and iterate as needed.

Multi-Modal Reference System

Seedance 2's core strength is its ability to combine multiple modalities. Each type of input serves a different creative purpose. Use @ syntax to assign roles.

📷 Image Reference

Upload images for character design, scene composition, style guide, or product shots. Up to 9 images per generation.

@image1 as first frame, character wears @image2 outfit, scene style like @image3

🎥 Video Reference

Upload videos for camera movement, action choreography, VFX templates, or editing rhythm. Up to 3 videos, total ≤15 seconds.

Reference @video1 camera movement and transition style, replicate @video1 action sequence

🎵 Audio Reference

Upload MP3 files for background music, rhythm sync, or voice reference. Up to 3 audio files, total ≤15 seconds.

@audio1 for background music, sync visual cuts to @audio1 beat

✍️ Text Prompt

Describe your scene, camera direction, character actions, and audio design in natural language. Be specific about what each @ reference should do.

@image1 character walks through @image2 scene, camera follows from behind, ambient rain sound

💡 @ Syntax Tips

• Type @ in the input to see all uploaded assets
• Be explicit about each reference's role: "@image1 as first frame" vs "@image1 for style"
• When using multiple references, label each clearly to avoid confusion
• You can reference a video's audio track without uploading a separate audio file
• Mixed input limit: 12 total files across all modalities

Video Editing & Extension

Seedance 2 doesn't just generate from scratch — you can edit existing videos, extend them, or insert new scenes between clips.

🎬 Character Replacement

Upload an existing video and new character images. Tell Seedance 2 to swap characters while preserving the original camera work, timing, and scene composition.

Replace the woman in @video1 with @image1 character, keep all camera movements and transitions

⏱️ Video Extension

Extend any video with smooth continuation. Describe what happens next and set the generation duration for the new portion only.

Extend @video1 by 10 seconds: the character walks toward the sunset, camera slowly pulls back

🔗 Scene Insertion

Insert a new scene between two existing video clips. Describe the bridging content and Seedance 2 creates a smooth transition.

Insert a scene between @video1 and @video2: a panoramic view of the city skyline at dusk

Practical Prompt Examples

Product Commercial

Inputs: 3 product images + 1 reference video (ad template) + 1 audio (upbeat music)

Create a 15s product commercial for @image1 handbag. Side details reference @image2. Surface texture reference @image3. Camera movement and transitions reference @video1. Background music: grand and cinematic. Show handbag details from multiple angles.

Camera & VFX Replication

Inputs: 1 character image + 1 reference video (camera work) + 1 scene image

@image1 character in @image2 corridor scene. Replicate all camera movements from @video1 — tracking shot from behind, then orbit to front face close-up. Character is running, out of breath, looking around nervously.

Music Beat Sync

Inputs: 6 scene images + 1 reference video (rhythm template)

@image1 through @image6 scenic shots, sync visual cuts to @video1 rhythm and transitions. Each image appears on the beat. Add dramatic lighting changes and camera movements matching the music energy. Dreamlike atmosphere with strong visual impact.

One-Take Tracking Shot

Inputs: 5 scene images (sequential locations)

@image1 @image2 @image3 @image4 @image5, one continuous tracking shot following a runner up stairs, through a corridor, across a rooftop, with the final shot overlooking the city. No cuts. Smooth camera movement throughout.

Tips & Best Practices

@ References

Always specify what each @ reference should do. "@image1 as first frame" is clearer than just "use @image1". When using many assets, double-check that you haven't mixed up images, videos, and audio references.

Asset Allocation

With a 12-file mixed limit, prioritize the assets that matter most. Upload the reference video for camera work first, then character images, then audio. Quality over quantity.

Duration Settings

Generation duration is 4–15 seconds. For video extensions, the duration setting controls the newly generated portion only. Want to extend by 10 seconds? Set duration to 10s.

One-Take Writing

For continuous tracking shots, describe the full camera path in sequence: "follows character up stairs → through corridor → onto rooftop." Seedance 2 handles spatial continuity automatically.

Describe Actions Clearly

Be specific about character actions and emotions: "character takes a deep breath, adjusts expression from tired to cheerful, then opens the door." Seedance 2 excels at nuanced emotion portrayal.

Iterate Gradually

Start with a simple prompt and one or two references. Once you get a good base result, add more modalities. Use video editing to refine, and extension to build longer sequences.

Ready to Create with Seedance 2?

Upload your images, videos, and audio — use @ references to take full control — and let Seedance 2 generate cinematic videos with built-in sound design.

Open Seedance 2 →