Seedance is ByteDance Seed's video generation model family, designed for high-fidelity text-to-video and image-to-video synthesis with multi-shot narrative continuity. It powers ByteDance's consumer-facing creative products such as Dreamina (CapCut / 即梦), Doubao, and Xiaoyunqe (小云雀), and is also exposed to developers through Volcengine.
Compared to general-purpose multimodal models, Seedance focuses specifically on the video generation domain, with emphasis on multi-shot consistency, motion realism, prompt adherence, and cinematic camera control.
Announced at the Volcano Engine FORCE conference on 2026-06-23 and launched globally on Dreamina on 2026-07-31. Key upgrades over 2.0:
- 30-second single-pass generation (with long-video extension mode up to ~3 minutes)
- Native 4K output (3840×2160) with 10-bit color depth
- Up to 50 multimodal references per request, including images, videos, audio, style references, 3D whitebox models, and greenscreen plates
- Region-level editing for localized modifications without regenerating the entire clip
- Maya / Blender plugins for production pipelines
Officially launched on 2026-02-12 (announcement). Previously in production behind Dreamina with 2K output support.
| Surface | Audience | Link |
|---|---|---|
| Dreamina (CapCut / 即梦) | End users / creators | https://dreamina.capcut.com/ |
| Doubao | End users (chat surface) | https://www.doubao.com/ |
| Xiaoyunqe (小云雀) | Creators (free Seedance 2.5 trial) | https://xyq.jianying.com/ |
| Volcengine Ark | Developers (API) | https://www.volcengine.com/ |
| Name | Description | Links | Fees |
|---|---|---|---|
| dexhunter/seedance2-skill | Skill to create better prompts for Seedance video generation. Covers Jimeng / Seedance input constraints, @ reference syntax, camera language, prompt structures, and templates for ads, dramas, MVs, and educational videos. Compatible with Claude Code, Cursor, Cline, and other agent tools. | Github  |
Free |
Seedance 1.0 reports results on:
- SeedVideoBench-1.0 — ByteDance Seed's internal video benchmark
- Artificial Analysis Video Arena — third-party blind evaluation arena
Numeric scores are not reproduced here; check the technical report and the arena leaderboard for current standings.
- Short-form video creation: Generate clips for Douyin / TikTok / Reels directly from prompts or storyboards
- Image-to-video animation: Bring static illustrations, product shots, or character art into motion
- Multi-shot storytelling: Generate narrative sequences with consistent characters across shots — Seedance's headline capability vs. single-shot competitors
- Pre-visualization & storyboarding: Quickly draft scene ideas before committing to traditional production
- Marketing & e-commerce: Generate product showcase videos at scale
Seedance sits alongside:
- KLING AI (Kuaishou) — strong on long-form coherence and motion control
- Wan2.6 (Alibaba) — Alibaba's competing video stack
- hailuoai (MiniMax) — known for cinematic output
- Dream Machine (Luma) / Runway Gen series / Pika — Western alternatives
Seedance's competitive positioning centers on native multi-shot narrative generation and tight integration with the Dreamina / CapCut editing workflow.
- ByteDance Seed — Seedance product page: https://seed.bytedance.com/en/seedance
- Technical report (Seedance 1.0): https://arxiv.org/abs/2506.09113
- ByteDance Seed blog (Seedance 2.0 launch announcement, 2026-02-12): https://seed.bytedance.com/en/blog
- Volcano Engine FORCE conference 2026 (Seedance 2.5 announcement, 2026-06-23): https://www.volcengine.com/
- Volcengine: https://www.volcengine.com/
- Dreamina: https://dreamina.capcut.com/