The Short Version
ByteDance officially shipped Seedance 2.0 on February 12, 2026. After a brief pause in March for legal review, the global API rollout went live through fal on April 9, and Dreamina Seedance 2.0 landed in CapCut across major markets.
For TeamDay users, none of that complexity matters: Seedance 2.0 is wired into the platform via ModelArk and fal. No API key, no waitlist, no gated rollout. Every agent with the video-generation skill can call it today.
Model Card
| Field | Value |
|---|---|
| Provider | ByteDance |
| Model | Seedance 2.0 |
| Released | Feb 12, 2026 (global API: Apr 9, 2026) |
| Modes | Text-to-video, image-to-video |
| Max resolution | 1080p |
| Single-shot length | up to 12 seconds |
| Multi-shot | Yes — character + scene continuity across cuts |
| Audio | Synchronized SFX and music |
| Access in TeamDay | Media MCP server (ModelArk + fal upstream) |
| Best for | Brand films, ad cuts, social video, product hero loops |
What's Genuinely New
The 1.0 → 2.0 jump is bigger than the version number suggests. Three things stand out:
- Multi-shot continuity. Seedance 1.0 made one shot at a time. 2.0 chains shots together with the same character, lighting, and scene logic — meaning you can ask for "a 30-second brand film with three cuts" and get an actual story, not three disconnected clips.
- Synchronized audio. SFX and music are generated alongside the video, timed to motion. No more dropping clips into a separate audio tool.
- Text-to-video that respects camera language. Phrases like "slow dolly-in," "handheld follow," "rack focus from foreground to logo" actually do what they say.
The combination is what makes 30-second branded ads tractable from a single prompt.
Example Prompts to Try
Brand film, 30s
"30-second brand film for a third-wave coffee roaster. Shot 1 (8s): slow dolly-in on roasted beans tumbling in a drum, warm amber light. Shot 2 (10s): a barista's hands pulling an espresso shot, shallow depth, milk steaming in the background. Shot 3 (12s): a customer's first sip, soft smile, cuts to logo: 'Drift Coffee — slow on purpose.' Mood: quiet, deliberate, no music — only ambient roastery sounds."
Product hero loop, 6s
"6-second seamless loop for a hero section. Subject: a matte-black wireless headphone rotating slowly on a polished concrete plinth. Light: single key from camera-left, deep shadows. Background: clean charcoal gradient. No text, no logo. Loop should be invisible at the seam."
Image-to-video, 8s
Input: product photo of a leather wallet on a wooden desk. Prompt: "8-second image-to-video. Camera: slow handheld push-in. Add subtle environmental motion — dust drifting in window light, page of an open notebook turning at the edge of frame. Hold focus on the wallet. End on a still."
Social ad, 9:16, 15s
"15-second vertical ad for a meal-kit subscription. Three quick cuts: (1) a busy parent opening a delivery box, (2) chopping vegetables in a sunlit kitchen with a child watching, (3) the family eating, laughing. Energetic but not frantic. Light, modern indie-pop instrumental. End frame: app icon and 'First box, $1.'"
These four cover the patterns we see most: brand film, hero loop, product close-up, and vertical social.
How It Compares
| Model | Released | Max resolution | Multi-shot | Audio |
|---|---|---|---|---|
| Seedance 2.0 | Feb 12, 2026 | 1080p | Yes (continuity) | Sync SFX + music |
| Sora 2 | Q1 2026 | 1080p | Limited | Separate model |
| Kling 2.5 | Q1 2026 | 1080p | Yes | Sync (basic) |
| Veo 3.5 | Q1 2026 | 4K | Yes | Sync |
Veo wins on raw resolution. Seedance 2.0 wins on integrated audio + multi-shot from a single prompt — which is what most brand-video missions actually need.
How to Use Seedance 2.0 in TeamDay
How it's wired
Seedance 2.0 doesn't run inside a coding harness — it lives behind the TeamDay media MCP server. Any agent — Claude Code, Codex, or Gemini CLI — that has the media MCP server attached can call video.generate, and the server routes the request to Seedance via ModelArk + fal upstream. Same MCP server hosts gpt-image-2, music generation, and voice — one connector, every visual model.
That means there's nothing harness-specific about video. A Claude Code agent and a Codex agent can both cut you a brand film through the same MCP tool.
1. Attach the media MCP server
In Agent Settings → MCP Servers, enable teamday/media. The agent picks up video.generate, image.generate, audio.generate, and voice.synthesize tools automatically. AI Video Studio ships with it pre-attached.
2. Set Seedance 2.0 as the video provider
Under Workspace Settings → Media MCP → Video provider, choose seedance-2.0. Every agent in the workspace using video.generate will route to Seedance unless overridden per agent.
3. Use it inside a multi-step mission
The mission pattern that gets the best results:
- Brief intake (Claude Code / Opus 4.7) — agent extracts the brand voice, format, length
- Storyboard (Claude Code / Opus 4.7) — agent drafts 3 shot-by-shot treatments
- Approve — you pick one in your shared space
- Generate (media MCP / Seedance 2.0) — agent produces the video at 1080p
- Deliver — agent posts the .mp4 + a teaser GIF
This is the Brand Film Drop mission template — duplicate it and point it at your brand brief.
4. Pair it with gpt-image-2 for thumbnails
Run gpt-image-2 first to lock the visual style and approve a hero frame. Then feed that hero into Seedance 2.0 in image-to-video mode. You get a tight visual through-line from still to motion.
When Not to Reach for Seedance 2.0
- Long-form (>60s) narrative video — chain multiple multi-shot sequences with editorial in between, or use Veo 3.5 for the longest shots.
- 4K deliverables — Seedance caps at 1080p; upscale in post or use Veo 3.5.
- Highly stylized animation (anime, 2D motion) — purpose-built models like Pika or Animatediff still lead.
TL;DR
Seedance 2.0 is the first video model that handles multi-shot brand films with synchronized audio from a single prompt. Inside TeamDay it's wired in natively — no API key, no waitlist — and pairs cleanly with gpt-image-2 for the still-to-motion loop and Opus 4.7 for the brief-to-storyboard step.