Video Generation with Seedance 2.0 and ComfyUI

AI video generation has evolved rapidly, and Seedance 2.0 from ByteDance represents a significant leap forward. Unlike traditional text-to-video models, Seedance 2.0 introduces a unique @ reference system that lets you combine images, video, and audio as inputs—creating videos with unprecedented consistency and control.

In this guide, we’ll explore how to integrate Seedance 2.0 with ComfyUI, covering installation, workflow setup, prompt engineering, and how it compares to alternatives like LTX-2 and Wan 2.2.

What is Seedance 2.0?

Seedance 2.0 is ByteDance’s multimodal AI video generator. It’s API-only (no local model weights) and excels at:

• Native 2K resolution output (up to 2048p)
• 4-15 second video clips with smooth motion
• Synchronized audio-video generation with lip-sync in 50+ languages
• @-tag reference system for combining images, video, and audio inputs
• Character consistency across multiple shots

The @ Reference System

The standout feature is the reference system. Instead of relying solely on text prompts, you can use:

@Image reference_photo.jpg
@Video style_clip.mp4
@Audio background_music.mp3

This allows for:

• Style transfer from reference videos
• Character consistency from uploaded images
• Audio-reactive video generation
• Multi-shot sequences with the same subjects

Integration with ComfyUI

Seedance 2.0 connects to ComfyUI via the Sjinn.ai API wrapper node. This requires a Sjinn.ai Pro+ subscription (100 credits per second of video).

Installation

Method 1: ComfyUI Manager (Recommended)

1. Open ComfyUI Manager
2. Search for “Seedance 2”
3. Install Cameraptor/seedance_2_Comfy_UI_Node-sjinn_Api-
4. Restart ComfyUI

Method 2: Manual Installation

cd ComfyUI/custom_nodes
git clone https://github.com/Cameraptor/seedance_2_Comfy_UI_Node-sjinn_Api-

Credential Setup

You need two credentials from Sjinn.ai:

1. API Key — For authentication
2. Session Token — For account-specific features

In ComfyUI, add these to your environment variables or settings:

SJINN_API_KEY=your_api_key_here
SJINN_SESSION_TOKEN=your_session_token_here

Note: The session token is separate from the API key and required for Pro+ features.

Workflow Guide

Basic Text-to-Video

Basic workflow: Prompt → Seedance 2.0 → Video Output

1. Add the Seedance 2.0 Node to your workflow
2. Connect a Text Input node with your prompt
3. Set resolution (default: 1080p)
4. Set duration (4-15 seconds)
5. Execute and wait for API response

Multimodal Workflow

For more control, use the @ reference system:

Multimodal workflow with @Image, @Video, and @Audio references

@Image portrait.jpg
@Video cinematic_motion.mp4
@Audio ambient_drone.wav

A woman walking through a foggy forest at golden hour, 
camera slowly pushing in, atmospheric and dreamlike

Cost Management

Seedance 2.0 uses a credit system:

Duration	Credits
4 seconds	400 credits
8 seconds	800 credits
15 seconds	1500 credits

Tips to conserve credits:

• Use shorter clips for testing
• Generate at lower resolutions during iteration
• Finalize prompts before generating long clips

Prompt Engineering Tips

Seedance 2.0 responds well to structured prompts. Here’s a recommended format:

Structure Formula

[SUBJECT] doing [ACTION] in [ENVIRONMENT], 
[CAMERA movement], [LIGHTING/atmosphere], [STYLE]

Example Prompts

Portrait Video:

@Image professional_headshot.jpg

A business executive walking through a modern glass office building, 
slow dolly in following the subject, bright natural lighting through floor-to-ceiling windows,
corporate documentary style, 4K quality

Cinematic Scene:

@Video blade_runner_reference.mp4

A cyberpunk city street at night with neon signs reflecting on wet pavement,
steady handheld tracking shot, moody rain-soaked atmosphere,
cinematic sci-fi aesthetic with volumetric fog

Audio-Reactive Music Video:

@Audio electronic_track.wav

Abstract geometric shapes morphing and pulsing to the rhythm of the music,
camera orbiting the central form, neon color palette with black background,
psychedelic visualizer aesthetic

Prompt Length Guidelines

• Optimal: 150-300 characters — Enough detail without overwhelming
• Maximum: 500 characters — Longer prompts get truncated
• Minimum: 50 characters — Too short lacks control

Multi-Shot Techniques

For multiple shots with consistent subjects:

1. Upload a reference image with @Image
2. Use the same reference across all prompts
3. Describe camera movements explicitly: “slow zoom”, “tracking shot”, “static wide shot”

Comparison: Seedance 2.0 vs LTX-2 vs Wan 2.2

Feature	Seedance 2.0	LTX-2	Wan 2.2
Access	API only	Local	Local
Resolution	Up to 2K	Up to 4K	480-720P
VRAM Required	None	32GB+	8GB+
Audio	Native sync	Native sync	None
Reference System	@ tags	Image refs	Limited
Cost	Pay/second	Free	Free
Best For	Consistency	Quality	Consumer GPU

When to Use Each

Use Seedance 2.0 when:

• You need character consistency across shots
• Audio sync is critical (music videos, dialogue)
• You don’t have high-end GPU hardware
• You want to reference existing videos for style

Use LTX-2 when:

• Maximum quality (4K) is priority
• You have a 32GB+ VRAM GPU
• You want full local control
• You need longer videos (up to 60s)

Use Wan 2.2 when:

• You’re on consumer hardware (8GB+ VRAM)
• Quick experiments without API costs
• Lower resolution is acceptable
• You want local inference

Troubleshooting

Issue	Solution
Authentication failed	Verify both API key AND session token
Video quality poor	Check resolution setting, use @Video reference
Credits not deducting	Session token may be invalid
Slow generation	API response time varies (30-60s typical)
Motion artifacts	Reference higher quality source videos

Best Practices

Iteration Workflow

1. Start with a text-only prompt to validate concept
2. Add @Image references for subject consistency
3. Add @Video references for style/motion
4. Add @Audio for synchronization
5. Generate final at maximum resolution

Camera Control

Be explicit about camera behavior:

• Static: “locked off camera, wide shot”
• Motion: “slow dolly in”, “handheld tracking”, “orbit around subject”
• Style: “tripod mounted”, “gimbal stabilized”, “drone flyover”

Audio Sync

For lip-sync or music videos:

• Generate video first, then add audio
• Or provide audio upfront for sync generation
• 50+ languages supported for lip-sync

Conclusion

Seedance 2.0 brings something unique to AI video generation: controlled consistency through multimodal references. While it requires a subscription and API access, the ability to combine images, videos, and audio as inputs makes it powerful for productions requiring multiple shots with consistent subjects.

For local alternatives, LTX-2 offers higher resolution output, and Wan 2.2 works great on consumer hardware. The right tool depends on your project needs and hardware availability.

Resources

• Seedance Official Documentation
• Sjinn.ai Guide
• ComfyUI Node Repository
• LTX-2 Comparison Article
• Wan 2.2 Video Generation Guide