The Complete Seedance 2.0 (Jimeng) User Manual 206

I. Seedance 2.0 Overview

Jimeng Seedance 2.0 is the next-generation AI video generation model launched by ByteDance. It adopts a dual-branch diffusion transformer architecture that can simultaneously generate video and audio. The biggest breakthrough of Seedance 2.0 is its upgrade from "text-to-video" to a multimodal video creation platform, taking creation from "generation" to "control". Feng Ji, the producer of "Black Myth: Wukong", praised it as "currently the strongest video generation model on the planet".

Core Features of Seedance 2.0

• Fast Output: Create a multi-shot video with original sound in about 60 seconds.

• Diverse Inputs: Supports text, image, video, and audio inputs simultaneously.

• Simple Operation: Automatic shot division and camera movement — no need to write complex prompts to describe camera motion.

• Audio-Visual Sync: Audio and visuals are aligned during generation, with accurate lip-sync.

II. Seedance 2.0 Access Entry

1. Jimeng AI (Primary Entry)

• URL: https://jimeng.jianying.com/

• Usage Requirements: Available directly for members (membership starts at approximately 69 yuan)

• Supported Platforms: Web version, Mobile App

2. Doubao App

• Invoke via prompts in Doubao: Use a prompt like "Use Seedance 2.0 to generate a Shaw Brothers-style fighting video", and Doubao can iteratively refine the prompt.

3. Xiaoyunque App

• Provides 3 free trial opportunities

III. Detailed Explanation of the Two Generation Modes

Mode 1: First/Last Frame Mode (Beginner-Friendly)

• Applicable Scenario: Only one first-frame image (or last-frame image) + text prompt

• Operation Steps:

• Select "AI Video" → Select "Seedance 2.0" → Select "Single Frame / First-Last Frame" mode

• Upload the first-frame image (optional: upload a last-frame image)

• Write a prompt describing the video content

• Set parameters (duration, aspect ratio, etc.)

• Click "Generate"

• Features: The simplest way to get started, ideal for quick experimentation

Mode 2: Full Reference Mode (⭐ Core Play)

• Applicable Scenario: Need to mix inputs of images + video + audio + text for precise control

• Operation Steps:

• Select "AI Video" → Select "Seedance 2.0" → Select "Full Reference" mode

• Upload reference materials (up to 12 files):

• Images: Up to 9 (characters, scenes, styles)

• Videos: Up to 3 clips (action, camera movement reference)

• Audio: Up to 3 clips (soundtrack, sound effect reference)

• Use the @material name syntax to specify the purpose of each material

• Write detailed prompts

• Generate video

IV. Detailed Explanation of @ Reference Syntax

Basic Reference Format

@图片1 作为首帧
@视频1 参考镜头语言
@音频1 用于配乐
@图片2 作为角色参考

Advanced Combination Usage

Correct Usage of @:

Set first frame and reference: @图片1 作为首帧，参考 @视频1 的镜头语言，@音频1 用于配乐

Set characters: @图片1 的女生作为主角，@图片2 的男生作为配角

Set camera movement: 完全参考 @视频1 的所有运镜效果和转场

Set scenes: 左边场景参考 @图片3，右边场景参考 @图片4

Set actions: @图片1 的人物参考 @视频1 中的舞蹈动作

Set voice: 旁白的音色参考 @视频1

💡 How to Trigger "@": Type the "@" character in the input box, or click the "@" button in the parameter tools bar

💡 Tip: File quantity is limited (up to 12); prioritize uploading materials that have the greatest impact on visuals and pacing

V. Prompt Writing Guide

Structure of an Excellent Prompt

[Scene Description] + [Character Action] + [Camera Movement] + [Atmosphere/Mood] + [Special Requirements]

Example Prompts

Basic Example:

樱花树下，一个女孩扭头看向一只猫，花瓣飘落，微风吹过她的头发。女孩摸了摸猫的头，对猫说"你好呀，小家伙"，最后画面定格在女孩脸上。

Complex Camera Movement Example:

The camera follows a man in black as he flees rapidly, with a group of people chasing behind. The camera shifts to a side tracking shot as the panicked figure crashes into a roadside fruit stand, scrambles back to his feet, and continues running, amid the chaotic sounds of the crowd.

Multimodal Reference Example:

@图片1 as the first frame, @视频1 referencing the camera movement style,
@音频1 as the background music. A martial arts hero practices swordplay in a bamboo forest, sword aura sweeping across as fallen leaves flutter. The camera slowly pushes in from a wide shot to a close-up, with a bleak, tragic atmosphere.

Prompt Optimization Tips

• Be specific: describe concrete actions, expressions, and clothing details

• Sense of scene: clearly define time, location, and lighting conditions

• Sense of camera movement: even without detailed description, the model will auto-storyboard, but you can specify special camera movement requirements

• Emotion words: use words like "bleak," "cheerful," "tense" to guide the atmosphere

VI. Detailed Parameter Settings

Basic Parameters

Advanced Settings

• First-frame control: upload the first image to define the video's starting frame

• Last-frame control: upload the final image to define the video's ending frame

• Reference strength: controls how much the AI learns from reference materials

VII. In-Depth Analysis of Core Features

1. Auto-Storyboarding and Auto-Camera Movement

• Breakthrough: no need to precisely describe complex camera instructions like "camera pans from left to right" or "wide shot followed by close-up."

• How it works: the model automatically plans storyboards and camera movements based on the plot, generating sequences with shot-size variations, camera motion, and spatiotemporal coherence.

• Effect: a simple prompt can produce results comparable to a professional director's camera work.

2. Comprehensive Multimodal Reference

Supported Files:

• Images ×9: character design, scene atmosphere, clothing details

• Videos ×3: camera movement style, action poses, rhythm reference

• Audio ×3: background music, sound effect style, voiceover reference

Advantages: pixel-level guidance of generation results, effectively mitigating the issue of excessive randomness in AI-generated content.

3. Native Audio-Video Synchronization

Capabilities:

• Precise matching of character lip movements with speech rhythm

• Facial expressions aligned with audio emotion

• Phoneme-level lip-sync across 8+ languages

• Environmental sound effects physically matching scene materials

• Value: no post-production voiceover needed—truly one-click generation of a complete video.

4. Multi-Shot Narrative Capability

Features:

• Consistent character appearance across multiple shots

• Continuous lighting and color tone

• Unified overall visual style

• Application: directly generates complete narrative segments with multiple shot transitions, without the need for complex workflow fixes.

VIII. Post-Production Editing and Optimization

Post-Generation Features

Download and Sharing

• Resolution: supports 1080p–2K HD output

• Format: standard video format, compatible with all platforms

• Copyright: generated videos come with full commercial usage rights, no attribution required

IX. Application Scenarios

1. AI Comic Dramas / Short Drama Production

Workflow:

• Use an LLM to break down the novel into a storyboard script

• Use Seedream 5.0 to produce character reference sheets (front view, side view, full body)

• Use Seedance 2.0 to batch-generate 5–15 second shots

• Leverage the multi-shot narrative feature to maintain character consistency

• Natively generate voiceovers and sound effects

2. E-commerce Advertising

• Rapid generation of product showcase videos

• Automatic switching between multi-angle shots

• Background music matched to product atmosphere

3. Film and TV Pre-visualization

• Quickly generate proof-of-concept videos

• Test different camera movement options

• Reduce pre-production shooting costs

4. Social Media Content

• Douyin/TikTok vertical videos

• Xiaohongshu recommendation videos

• YouTube horizontal content

X. Usage Limitations and Precautions

⚠️ Current Limitations

• Real-person content restriction: real-person images or videos are currently not supported as primary references

• Jimeng Web and Xiaoyunque explicitly indicate that real human faces are not supported as reference materials

• In Jimeng App and Doubao App, if you need to use a real-person likeness, you must complete identity verification by recording your own likeness and voice before creating a digital avatar

Generation Suggestions

• Avoid overly simple prompts: for facial emotion expression of animated characters, overly simple prompts may result in stiff expressions

• Chinese text: there is still room for improvement in rendering Chinese text

• Credit management: keep an eye on Jimeng membership credit consumption and plan your generation count wisely

XI. Quick Start Checklist

• Visit the Jimeng AI official site (jimeng.jianying.com) and log in

• Confirm your membership status (or claim a free trial on Xiaoyunque)

• Prepare reference materials (images/videos/audio)

• Choose a generation mode (First & Last Frame or All-Powerful Reference)

• Write a detailed prompt (scene + action + atmosphere)

• Use the @ syntax to specify the purpose of materials (All-Powerful Reference mode)

• Set video parameters (aspect ratio, duration, style)

• Generate and preview the result

• Use post-production features to optimize (if needed)

• Download the HD video

Jimeng Seedance 2.0 Official User Manual:https://bytedance.larkoffice.com/wiki/A5RHwWhoBiOnjukIIw6cu5ybnXQ