Seedance 2.0 Advanced Techniques

Master multimodal inputs, character consistency, and creative workflows to unlock the full potential of Seedance 2.0 video generation.

Multimodal Input System

Seedance 2.0 supports four input modalities — images, videos, audio, and text — that can be freely combined for richer expression and more controllable generation. The @tag reference system lets you point to specific uploaded assets directly inside your prompt so the model knows exactly which input to use for which purpose.

The @Tag Reference System

When you upload images, videos, or audio files, Seedance 2.0 assigns them numbered tags. Reference these tags in your prompt text to tell the model how each asset should influence the output.

@Image1 – @Image9 (up to 9 images) — controls visual style, character appearance, scene composition, product details, and lighting reference. Usage: "Use the character design in @Image1 and the background style of @Image2."

@Video1 – @Video3 (up to 3 clips) — controls motion rhythm, camera movements, action choreography, special effects, and video continuation. Usage: "Reference the camera movement and rhythm in @Video1."

@Audio1 – @Audio3 (up to 3 audio files) — controls music beat synchronization, mood and pacing, voice-over reference, and ambient sound. Usage: "Synchronize the video rhythm to the beat of @Audio1."

Example: Multimodal Prompt

Reference the character design and outfit in @Image1. A young woman in a flowing red dress walks confidently down a neon-lit city street at night. Reference the smooth tracking camera movement and rhythm from @Video1. Synchronize the motion pacing to the beat of @Audio1. Cinematic film quality with shallow depth of field, rain-slicked reflections, and vibrant neon color cast.

Tip: You can mix and match any combination of inputs. Use an image for appearance, a video for motion reference, and audio for rhythm — all in a single prompt. Clearly state which aspect each asset controls for best results.

Character Consistency

Maintaining consistent character appearance across multiple generated clips is one of the most important skills for professional-quality video production. Seedance 2.0 offers comprehensive consistency improvements — from facial features and clothing to font details and overall style retention.

Rules for Consistent Characters

1. Use Consistent Character Descriptions. Write the exact same character description in every prompt. Include specific details like hair color, clothing, accessories, and distinguishing features. Avoid vague terms — be precise with "a young woman with shoulder-length auburn hair, wearing a navy trench coat and silver pendant necklace" rather than just "a woman."

2. Reference the Same Image Across Generations. Upload a character reference image and use @Image1 in every prompt for that character. This anchors the model to the same visual identity across all shots, ensuring facial features, body proportions, and outfit details remain consistent.

3. Keep Lighting and Style Consistent. Use the same lighting setup and visual style keywords across all shots in a sequence. Switching from "golden hour warm tones" to "cool blue moonlight" between shots will make the character look different even with the same description. Lock in one lighting anchor per scene.

4. Use Similar Camera Angles. Dramatic angle changes can alter perceived proportions and features. When consistency is critical, keep camera angles within a natural range (e.g., eye-level to slight low-angle) and avoid extreme overhead or worm's-eye views between consecutive shots.

Example: Consistent Character Prompts

Shot 1:

Reference the character in @Image1. A young woman with shoulder-length auburn hair, wearing a navy trench coat and silver pendant necklace, walks slowly through a rain-soaked Paris street at dusk. Slow dolly-in from medium shot to close-up. Cinematic film quality, warm golden hour sunlight with soft amber lens flare.

Shot 2 (same character, different scene):

Reference the character in @Image1. A young woman with shoulder-length auburn hair, wearing a navy trench coat and silver pendant necklace, sits at a cafe table and slowly raises a coffee cup with a thoughtful expression. Static medium shot, shallow depth of field. Cinematic film quality, warm golden hour sunlight with soft amber lens flare.

Multi-Shot Storytelling

Create narrative continuity across multiple generated clips by establishing a character first, then building a visual story shot by shot. Each clip should flow naturally into the next, maintaining character identity, visual style, and emotional arc.

Step-by-Step Workflow

Step 1 — Establish Your Character. Start by generating or uploading a clear character reference image. This becomes your visual anchor for every subsequent shot. Describe the character in detail once and save that description for reuse.

Step 2 — Generate the First Shot (Establishing). Create the opening shot that introduces the character and setting. Use a wider camera angle to establish the scene and context.

Step 3 — Generate the Second Shot (Development). Move the story forward with an action or event. Shift camera angle slightly for visual variety while keeping the character description and style identical.

Step 4 — Generate the Third Shot (Climax / Resolution). Close the narrative arc with a dramatic or emotional moment. Use a tighter shot or dramatic camera movement to heighten impact.

Example: 3-Shot Story — "The Discovery"

Shot 1 — Establishing: The Explorer Arrives

Reference the character in @Image1. A rugged explorer in a weathered leather jacket and wide-brim hat stands at the entrance of an ancient overgrown temple deep in a misty jungle. Static wide shot capturing the full scale of the temple entrance. Cinematic film quality, dappled forest light filtering through leaves with green-tinted ambient glow and volumetric mist.

Shot 2 — Development: Into the Temple

Reference the character in @Image1. A rugged explorer in a weathered leather jacket and wide-brim hat walks slowly forward through a dark stone corridor illuminated by flickering torchlight, dust particles floating in the air. Tracking shot following alongside the subject maintaining steady distance. Cinematic film quality, warm candlelight creating dancing shadows on ancient carved walls.

Shot 3 — Climax: The Discovery

Reference the character in @Image1. A rugged explorer in a weathered leather jacket and wide-brim hat slowly raises gaze toward a massive golden artifact glowing with ethereal light in a vast underground chamber. Slow dolly-in from medium shot to intimate close-up on the explorer's awestruck face. Cinematic film quality, dramatic volumetric light rays cutting through atmospheric haze with golden warm tones.

Beat-Sync Video

Seedance 2.0 enables precise synchronization between video motion and music beats. Upload an audio track using the @Audio tag and the model will align motion dynamics, camera transitions, and visual rhythm to the tempo and energy of your music.

How It Works

1. Upload Your Audio. Upload a music track or audio clip — a few seconds is enough. The model analyzes the beat structure, tempo, and energy dynamics automatically.

2. Reference with @Audio. Use @Audio1 in your prompt and explicitly tell the model to synchronize motion to the beat. Specify which aspects should follow the rhythm.

3. Add Rhythm Keywords. Include keywords that describe how the motion should relate to the music: synchronized to the beat, rhythmic motion, moves on each beat drop, pulsing with the rhythm, tempo-matched movement, beat-driven transitions, music-synced camera cuts, dynamic rhythm matching.

4. Describe the Visual. Complete the prompt with your subject, scene, camera, and style as usual. The beat-sync layer works on top of standard prompt structure.

Example: Beat-Sync Prompt

Synchronize all motion to the beat of @Audio1. A dancer in a flowing black costume performs powerful rhythmic movements on a dark stage with dramatic spotlights, each motion hitting precisely on the beat drop. Dynamic camera cuts synchronized to the rhythm, alternating between wide and close-up angles. Cinematic high-contrast lighting with volumetric haze and vibrant stage colors pulsing with the music.

Video Extension

Extend existing video clips seamlessly using the @Video tag. Instead of regenerating an entire video from scratch, you can use an existing clip as input and add seconds of natural continuation — preserving character identity, scene context, motion trajectory, and visual style.

Tips for Seamless Continuation

Match the Ending State. Describe the continuation from exactly where the original clip ends. If the character was mid-stride, continue that stride rather than starting a new action abruptly.

Preserve Visual Style. Use the same lighting, color grading, and style keywords as the original video. Switching styles mid-extension creates a jarring visual discontinuity.

Keep Camera Consistent. Continue the same camera movement and angle from the end of the reference clip. Sudden camera jumps break the illusion of a continuous take.

Use Gentle Transitions. If introducing a new action, use transitional language like "then slowly begins to" or "gradually transitions into" for natural flow.

Example: Video Extension Prompt

Continue the video from @Video1. The young woman continues walking forward along the rain-soaked city street, then gradually slows to a stop and turns her head toward a glowing shop window. The camera continues its smooth tracking movement, slowly transitioning into a gentle dolly-in as she pauses. Maintain the same cinematic film quality, warm golden tones, and shallow depth of field with rain-slicked reflections.

Tip: Always start your extension prompt with "Continue the video from @Video1." This tells the model to treat the uploaded clip as the beginning of a continuous take rather than just a style reference.

Audio Generation

Seedance 2.0 features native sound generation that produces realistic audio matched to video content. From environmental sound effects and musical scoring to character dialogue with phoneme-level lip synchronization, the audio system supports natural voice generation across 8+ languages.

Audio Capabilities

Sound Effects. Generate context-aware environmental audio: rain, footsteps, explosions, wind, traffic, and more. Describe the sounds you want and the model generates them in sync with the visuals.

Phoneme-Level Lip Sync. Character dialogue is generated with precise lip synchronization at the phoneme level. Mouth movements match speech naturally, creating believable talking characters.

8+ Language Support. Native voice generation supports English, Chinese, Japanese, Korean, Spanish, French, German, Portuguese, and more.

Sound Generation Keywords

character speaks with clear voice, precise lip-sync, ambient rain pattering, footsteps echoing, wind howling softly, orchestral background score, natural voice dialogue, realistic sound effects

Example: Dialogue with Lip-Sync

A young journalist in a gray blazer sits across from the camera in a warmly lit studio, speaking directly to the viewer with a clear natural voice and precise lip-sync: "The discovery changed everything we thought we knew about the ancient civilization." Medium close-up, shallow depth of field. Professional three-point studio lighting with clean key light and soft fill. Subtle ambient room tone in the background.

Motion Transfer

Use reference videos as motion sources to transfer complex choreography, camera language, and movement rhythms to new subjects and scenes. The @Video tag lets you specify exactly which video to use as a motion template while freely changing the character, environment, and visual style.

How Motion Transfer Works

1. Upload a Reference Video. Choose a video clip that contains the motion, camera movement, or choreography you want to reproduce. The model analyzes its movement rhythm, camera language, and visual structure.

2. Describe What to Transfer. Be specific about which aspects of the reference video to use — the camera movement, the body motion, the rhythm, or all of the above.

3. Define Your New Subject. Replace the original subject with your own character and scene. The transferred motion will be applied to the new subject while respecting their proportions and context.

4. Combine with Other Inputs. Stack motion transfer with image references for character appearance and audio for beat synchronization. The multimodal system lets you compose all layers together.

Example: Motion Transfer Prompt

Reference the camera movement and action rhythm from @Video1, and the character design from @Image1. A warrior in ornate samurai armor executes the same combat sequence from the reference video in a bamboo forest at dawn. Replicate the dynamic camera transitions and motion pacing precisely. Cinematic epic style, dappled forest light filtering through bamboo leaves with golden morning mist.

Tip: For best motion transfer results, clearly separate what you want to transfer ("the motion choreography from @Video1") from what you want to change (new character, new scene, new style). The more explicit you are about which elements come from the reference versus your prompt, the more accurate the result.