Dreamina Seedance 2.0 series prompt guide

The Dreamina Seedance 2.0 series (Seedance 2.0 series) model natively supports joint audio and video generation, with excellent semantic understanding and multimodal interaction capabilities. This article introduces how to write prompts for the Seedance 2.0 series model and related tips, helping you use the model more efficiently to generate high-quality videos that meet your needs.

Note
All visual (image, video) and audio assets presented in this guide are autonomously generated by the Seedance/Seedream series visual generation models.

Basic formula

The Seedance 2.0 series model supports referencing multimodal assets such as videos, images, and audio at the same time. It can accurately lock in characteristics such as character appearance, action effects, visual style, and voice-over timbre, greatly lowering the threshold for writing prompts. Based on this advantage, we can guide the model with a simple basic formula, using the features of multimodal assets to quickly generate videos that meet specific requirements.
Reference-based video generation can be subdivided into three types of tasks: multimodal reference, editing videos, and extending videos. You can choose the basic prompt formula according to the task type.

Multimodal reference

Extract some elements from the asset (such as subject, style, scene, and sound effects) to generate a brand-new video.

• Applicable scenarios : Action transfer, subject reuse, atmosphere reference, etc.
• Recommended patterns :
  Image reference : Reference <Subject_N> in <Image_N> to generate...
  Video reference : Reference <Action/Camera_movement/Style/Sound_effect> in <Video_N> to generate...
  Audio reference : Reference the timbre in <Audio_N> to generate...

Video editing

Make partial or global modifications based on the original video. Parts not mentioned remain unchanged by default.

• Applicable scenarios : local replacement, subject removal, attribute modification, etc.
• Recommended patterns :
  Add elements : Clearly describe <Element_Features> + +
  Modify elements : Strictly edit <Video_N>, and modify <Original_Characteristic> in it to <New_Characteristic>
  Delete elements : Specify the elements that need to be deleted. For elements that should remain unchanged, emphasize them in the prompt for better performance.

Video extension

Continue the original video along the time dimension, requiring the audio-video style, subject, and narrative to remain consistent.

• Applicable scenarios : continuing the plot, extending actions, completing clips, etc.
• Recommended patterns :
  Extend video : Extend <Video_N> forward/backward to generate...
  Track completion : <Video_1> + <Transition_Description> + followed by <Video_2> + <Transition_Description> + followed by <Video_3>

Note
For edit / extend video tasks, directly use "<Video_N>" to refer to the video. Do not use "reference <Video_N>", to avoid being incorrectly identified as a reference task.

Combined tasks

The above three types of tasks can also be used in combination.

• Applicable scenarios : reference one asset and edit another asset.
• Recommended sentence patterns :
  Reference [Reference_Dimension] of <Image/Video_N>, strictly edit <Video_X>, [Specific_Edits]

Advanced formula

Seedance 2.0 is essentially a multimodal AI director: it reads your text prompts, images, videos, and audio simultaneously, and internally breaks them down into two dimensions, the "spatial layer" (what is in the frame) and the "temporal layer" (how things change over time), to understand and generate visuals.
Therefore, a good prompt is not simply "copywriting-style description", but an "engineering-style instruction": who, in what scene, doing what action, how the camera moves, and in what chronological order events occur are delivered respectively to the spatial layer and the temporal layer. The specific formula is as follows:

Advanced prompt formula: precise subject + action details + scene/environment + lighting & color tone + camera movement + visual style + image quality + constraints

In simple terms, first lock in "who" is "doing what", then explain "where" and "what atmosphere", then tell the model "how to shoot", and finally tighten the result with style, image quality, and constraints. The following sections break down each element in detail.

1. Define the subject

In the actual referenced asset, one image often contains multiple subjects. To precisely reference a specific object in the asset, you must clearly define the subject. The subject can be a person, prop, scene, etc.

Note

• Each time a subject is involved, it must be explicitly referred to to avoid omission. The following two usages are supported:
• For simple scenarios with undefined subjects, each time the subject is mentioned, use <Subject_N>@<Image_N> to emphasize the binding relationship between the subject and the asset. For example: Zhang San@Image 1.
• For scenarios where subjects have been defined in advance, each time the subject is mentioned, the same label should be used. For example: define the tall man in Video 1 as police officer , and define the other short man as thief . In subsequent descriptions, when referring to the tall man, consistently use " police officer "; when referring to the short man, consistently use " thief ".
• When using the asset library (Asset ID), you still need to use <Image/Video_N> to refer to the subject. Because the model cannot directly associate the Asset ID with the reference content, you must not directly use the Asset ID instead of <Image/Video_N>.
• Keep descriptions as concise as possible, avoid redundancy, and avoid semantic conflicts (such as contradictory characteristics for the same subject).
• It is recommended to express spatial relationships through reference images first, reducing complex text descriptions.

2. Shot sequencing

The model's internal modeling decouples space and time. Therefore, the ideal prompt form for a complex video is a timeline-based storyboard: break the video into several shots, and dynamically describe each shot in the order events occur: who + where + doing what + how the camera moves.
Suggestions
Use shot order, write a simple "Shot 1 / Shot 2 / Shot 3" storyboard for each segment of the video, and then merge them into a complete prompt.

• Negative example : "A man runs nervously down the street, and the scene feels very cinematic."
• Positive example :
  Shot 1: Side shot of a street alley; the man slowly starts running, with a sense of rapid breathing.
  Shot 2: The man knocks over a fruit stand; the camera shakes quickly and gives a close-up of the man's frightened face.
  Shot 3: The man climbs over a low wall and disappears; the camera slowly pulls back and freezes on the empty street.
  Rules
• Use identifiers such as Shot 1, Shot 2, and Shot 3 to organize content in the order events occur (primary first, secondary later). Do not impose strict limits on the duration of each segment; prioritize allowing the model to naturally generate the pacing based on the plot.

Tip
The model's support for precise timing (such as 0–3 seconds) is unstable, and forcibly limiting duration may lead to abnormal generation results.

• It is recommended to organize each shot according to the following logic:

1. Camera movement or shot transition method : such as "slowly push in from a wide shot," "fixed camera position," "cut to..." and so on.
2. Subject actions and expressions : describe the key actions and changes in expression of the core character / object.
3. Position or spatial changes : explain the scene, position, or spatial relationship where the subject is located.
4. Audio information : describe the sound effects, voices, background music, etc. corresponding to the shot.

3. Action description requirements

• Body-movement refinement + degree quantification
  Actions should be specific to body parts such as hands, legs, head, shoulders, and back, while also supplementing descriptions of range, speed, and force ;
  Examples : slowly raise a hand, quickly turn the head, push hard off the ground, slightly lower the head.
  Prioritize slow, gentle, continuous small movements
  Prioritize slow, gentle, coherent subtle movements, and try to avoid high-burst, large-dynamic actions such as sprinting, big jumps, and violent rolls.
  Examples : walk slowly, gently raise a hand, slightly lower the head, sit down naturally with the motion
• Supplement transitions between actions
  Specify the inertia and continuity between preceding and following actions to ensure the on-screen movement is coherent and natural.
  Examples : use the inertia of turning around to naturally raise a hand, naturally transition from a pause into raising a hand.
• Concrete external expression of emotions
  Use specific physical details to express emotions, replacing abstract words such as "very sad" and "extremely angry."

4. Camera movement prompts

The model has a strong understanding of camera movement terms, so you can directly use standard camera movement terminology, such as "medium shot, close-up, wide shot, slow push-in, smooth lateral tracking, fixed shot." For more, see Lens language.

Note
Try to specify only 1 type of camera movement in a single shot. Do not require push, pull, pan, and move at the same time, as this will increase image instability.

5. Image quality, style and constraint words

Image quality, style, and constraint words are key to controlling video generation results. They can define creative boundaries for the model, unify image quality and artistic tone, and avoid visual flaws and random deviations, making them necessary configurations for ensuring stable and compliant final output.
1 Image quality
Define image clarity, detailed textures, and lighting quality to improve the basic image quality of the final video.
Examples : HD, rich details, cinematic texture, natural colors, soft lighting
2 Style
Set the overall art style and visual tone to unify the artistic atmosphere of the image.
Examples : cyberpunk cool blue-purple tone, retro film, fresh Japanese style
3 Constraint words
Constraint words are very important. They can effectively avoid visual flaws, deformities, breakdowns, and unreasonable elements, and constrain the generation boundaries and stability.
Common constraint word templates :
**Avoid generating subtitles : **"keep it subtitle-free," "avoid generating any text or subtitles"
Avoid generating a Logo : "do not generate a logo"
Avoid generating a watermark : "do not generate a watermark"

Other tips

Text generation

Seedance 2.0 series models support generating common text. The model can automatically match an appropriate style and color based on the context, and also supports specifying the color, style, appearance method, appearance timing, and appearance position of text in the prompt. When writing, please prioritize common characters and avoid rare characters and special symbols to ensure the best presentation. Currently, scenarios such as ad slogans, subtitles, and speech bubbles are supported. For specific writing methods and cases, see Text generation.

Video extension vs segmented stitching

• Continuous long take (video extension) : suitable for "dialogue scenes" within a single scene, such as long conversations, emotional progression, and movement along a single path, to achieve an immersive and coherent one-shot effect.
• Scene / action turning points (segmented stitching) : suitable for plot turns or complex, fast-paced "action scenes," such as chases, fights, montages, etc. Independent clips can be generated and then edited together to ensure rhythm and visual impact.
  In actual production, the two methods are usually combined. For example, first use extension to generate a coherent conversation, then stitch in empty shots or transition clips, balancing immersion with rhythmic variation.

Asset configuration strategy

Assets are usually divided into four "functional roles":

1. Character anchoring: lock the character appearance
2. Scene tone-setting: lock the environment and style
3. Camera movement reference: lock the shot language and action rhythm
4. Rhythmic atmosphere: use audio to control emotion and timbre
   Recommended configuration (4-5 assets in total): 1-2 character images (facial close-up / full body) + 1 scene image + 1 camera movement video + 1 audio clip.

Tip
It is not recommended to use the full asset limit. Too many assets will make it difficult for the model to judge feature priorities, which can easily cause issues such as style conflicts, blurry subject identification, and generated results deviating from expectations.