技能 seedance-v2
🎬

seedance-v2

低風險

Generate Cinematic AI Video with Seedance 2.0 Pro

也可從以下取得: agentspace-so,runcomfy-com

Creating professional video content requires expensive equipment and expertise. Seedance 2.0 Pro on RunComfy enables AI-powered cinematic video generation with native lip-sync, multi-modal references, and production-ready output in seconds.

支援: Claude Codex Code(CC)
🥉 72 青銅
1

下載技能 ZIP

2

在 Claude 中上傳

前往 設定 → 功能 → 技能 → 上傳技能

3

開啟並開始使用

測試它

正在使用「seedance-v2」。 Generate a 6-second video of a barista explaining today's special in a warm friendly tone. Use the character from character-ref.jpg. Medium close-up, slow push-in, warm cafe lighting.

預期結果:

A 6-second cinematic video showing a barista (matching the reference image) in warm cafe lighting, naturally lip-synced as they explain today's special offer with a slow camera push-in.

正在使用「seedance-v2」。 Create a 10-second 9:16 video where the spokesperson delivers the new product announcement. Reference character from brand-spokesperson.jpg. Calm professional tone, office setting.

預期結果:

A 10-second vertical video featuring the brand spokesperson (consistent with the reference) delivering the announcement with natural lip-sync in a professional office environment.

安全審計

低風險
v1 • 6/1/2026

All 89 static findings are false positives. The 'Ruby/shell backtick execution' flags are markdown code fences containing documentation examples, not executable code. 'Weak cryptographic algorithm' flags are false positives on 'seed' as a video generation parameter. 'System reconnaissance' flags are false positives on documentation text. Network and filesystem references are legitimate RunComfy API endpoints and token storage paths documented in the skill. This is a documentation-only skill for a video generation CLI with no malicious intent.

1
已掃描檔案
182
分析行數
0
發現項
1
審計總數
未發現安全問題
審計者: claude

品質評分

38
架構
100
可維護性
87
內容
50
社群
90
安全
91
規範符合性

你能建構什麼

Marketing and Advertising Teams

Create brand-consistent spokesperson ads and social media content with natural lip-sync. Use image references to maintain brand identity across multiple videos.

Content Creators and Filmmakers

Generate cinematic previsualization sequences for film projects. Combine multiple reference media types to establish visual continuity before production.

Multilingual Brand Narratives

Generate brand-consistent video content in multiple languages by maintaining character image references while translating the text prompt.

試試這些提示

Basic Video Generation 
Generate a 5-second video of [describe your scene, action, and mood]. Medium close-up, natural lighting.
Lip-Synced Character Video 
Generate an 8-second video using [describe character from image reference]. [Describe the action and setting]. [Describe the speaker's tone, e.g., warm and friendly]. Medium close-up. 9:16 aspect ratio.
Multi-Modal Cinematic Scene 
Generate a 10-second video. Subject from image 1 performs [describe action] in [setting from video 1]. Voice tone matches audio 1. [Additional cinematic direction]. Use the locked shot style from the reference video.
Brand-Consistent Multi-Language Content 
Use the character from [brand-character.jpg]. Generate a [duration]-second video in [target language] where the spokesperson says: [translated message]. [Describe tone and style]. Maintain brand lighting and setting from reference.

最佳實務

  • Put stable identity elements (faces, logos, costumes) in image_url references instead of describing them in text prompts to avoid visual drift
  • Use plain language for camera direction (e.g., "medium close-up", "slow push-in", "handheld follow") and combine multiple directives for complex shots
  • Match aspect ratio of reference media to your intended output to avoid unwanted cropping and framing issues

避免

  • Describing a character's face in detail in the text prompt instead of using an image reference, which causes identity drift
  • Mixing radically different aesthetic references (e.g., watercolor style with photorealistic) in the same generation, which confuses the model
  • Requesting video durations longer than 15 seconds, which will fail with a 422 error - instead, segment into multiple sequential calls

常見問題

What is the difference between Seedance 2.0 Pro and HappyHorse 1.0?
Seedance 2.0 Pro excels at multi-modal reference support and native lip-sync. HappyHorse 1.0 currently ranks higher in blind-vote video quality. Choose Seedance for lip-synced dialogue with character consistency; choose HappyHorse for best overall video quality.
How does lip-sync work on Seedance 2.0 Pro?
Seedance generates synchronized speech and lip movement in a single pass when generate_audio is enabled. You can guide the voice tone using descriptive text like 'warm friendly conversational' or 'calm instructional'.
Can I use my own audio track with lip-sync?
For audio-driven lip-sync from your own audio file, use Wan 2.7 with the audio_url parameter instead. Seedance generates audio in-pass rather than syncing to provided audio.
What aspect ratios does Seedance 2.0 Pro support?
Seedance supports: adaptive, 16:9 (landscape), 9:16 (vertical), 4:3, 3:4, 1:1 (square), and 21:9 (ultrawide cinematic).
How do I get consistent character identity across multiple videos?
Upload your character reference to image_url. The model maintains identity from the reference image. Avoid describing the character's appearance in the text prompt to prevent drift.
What happens if my reference media doesn't match the output format?
The model may crop or reframe your references to match the target aspect ratio. For best results, use reference media with the same aspect ratio as your intended output.

開發者詳情

作者

doany-ai

授權

MIT

儲存庫

https://github.com/doany-ai/skills/tree/main/seedance-v2/

引用

main

檔案結構

📄 SKILL.md