LESSON 1 · GETTING STARTED

Introduction: what Seedance 2.0 is and what it can do

Seedance 2.0 is an advanced video-generation neural network from ByteDance, running on the BytePlus ModelArk platform. It turns your ideas into finished video clips, drawing on four sources at once: text, images, video and audio. In this lesson you'll get an overall map of the model's capabilities — so that later you can steer each of them deliberately.

What Seedance 2.0 is

In short: you describe a scene in words and, if you like, add references (images, video fragments, sound), and the model synthesizes a coherent video clip with motion, light and camera. The key difference from "animating a single picture" is that Seedance 2.0 understands the intent of the scene as a whole: who's in the frame, what's happening, where, how the camera moves and in what style everything is shot.

🔑 Key idea
Seedance 2.0 is a multimodal model. It accepts several types of input at once and decides on its own how to combine them into a single clip. The more precisely you divide the tasks among text, image, video and audio, the more predictable the result.

The four modalities: who's responsible for what

This is the single most important idea of the whole course. Each modality has its own "zone of responsibility." Don't try to specify the rhythm of gestures with words or describe mood with a picture — every input has its own strength.

📝
TEXT

Builds the world

Text sets the space: the hero's appearance, the environment, the mood, the style and the quality. The best modality for everything that's about "how it looks and how it feels."

🖼️
IMAGE

Locks in identity

An image nails down exactly how a character, object or scene looks. An uploaded picture can become literally the first frame of the clip.

🎞️
VIDEO

Sets the motion

A video reference is responsible for time: timing, the rhythm of gestures, the trajectory and character of camera movement. The things that are hard to put into words.

🔊
AUDIO

Shapes rhythm and sound

Audio sets the scene's rhythm and its soundtrack — from the pace of the edit to the way motion syncs with sound.

Remember a simple rule: text — for the spatial (appearance, mood, style), video — for the temporal (timing, rhythm, motion), image — for identity, audio — for rhythm and sound. We'll return to this division in every lesson.

Input modes: how you can "talk" to the model

Seedance 2.0 supports several working modes. You pick the one that fits your task — from pure text to a complex multimodal assembly.

Text → video

You write only a prompt. The model invents the shot entirely from your description. The basic and most flexible mode.

Image → video

A single uploaded image becomes literally the first frame, and the model "brings the scene to life" from there following your description.

First + last frame

Interpolation: you set the starting and final frames, and the model smoothly builds the transition between them.

Multimodal reference

Up to 9 images, up to 3 videos and up to 3 audio together: character, scene, style, motion and rhythm are assembled into a single clip.

Video editing (V2V)

You feed in a finished video and ask to change the costume, background or lighting while preserving the original motion and timing.

Multi-shot scenes

A single prompt describes several consecutive shots with different actions and camera movement in each.

The model's technical capabilities

Before diving into prompts, it's useful to know the limits the model works within and what you can tune during generation.

480–1080p
resolution (natively up to 4K when detail is requested)
4–15 sec
length of a single clip
7+
aspect ratios, including auto
2
versions: standard and Fast
CapabilityWhat it gives you
Resolution480p, 720p or 1080p. Natively the model can deliver up to 4K if you ask for texture and face detail in the prompt.
LengthClips of roughly 4 to 15 seconds. For longer stories, multi-shot is used.
Aspect ratio16:9, 9:16, 1:1, 4:3, 3:4, 21:9 and adaptive/auto — for any format.
Model versionsSeedance 2.0 — maximum quality. Seedance 2.0 Fast — faster and cheaper, slightly lower quality: perfect for drafts and tests.
✅ There's nothing special to install
This course teaches the Seedance 2.0 model itself, not any single interface. All the principles — the prompt formula, the input modes, the reference labels — work the same way in any generation tool where the model is available. Once you've mastered the model's logic here, you'll be able to apply it anywhere.

What you'll master in this course

Next we'll go step by step through everything you need to work confidently with the model:

  • How generation works. What happens "under the hood" and how the model interprets your request.
  • Your first generation. We launch the simplest text-based clip and break down the result.
  • Model parameters. Resolution, length, aspect ratio and seed for repeatability.
  • The prompt formula. Subject + Action + Environment + Camera + Lighting + Style + Quality — the foundation of everything.
  • The three input modes. Text→video, image→video and first-and-last-frame interpolation.
  • Multimodal references. How to combine images, video and audio and refer to them with labels.
  • Consistency. How to keep the same hero and style across generations.
  • Multi-shot. Assembling several shots with different actions and camera in one prompt.
  • Realism and V2V. Photorealistic imagery and editing a finished video.
  • Workflow. How to iterate fast and not burn attempts for nothing.
  • Ready-made examples. A breakdown of working prompts you can use as a starting point.

Ready? In the next lesson we'll peek "under the hood" and break down exactly how the model turns your text and references into frames.

Try Seedance 2.0 yourself

Generate your first clip in ZenCreator — Seedance 2.0 is available in the Image-to-Video tool.

Start Creating