Seedance 2.0: ByteDance's video model that scandalized Hollywood and leads global benchmarks
ByteDance launched Seedance 2.0 in February 2026 and within 48 hours had generated clips so realistic — Tom Cruise fighting Brad Pitt — that Hollywood responded with cease-and-desist letters. The global launch was paused, filters were redesigned, and today the model is available on CapCut and Dreamina — with the world's highest Elo in video generation.
Elo 1,269
text-to-video — #1 worldwide
Elo 1,351
image-to-video — #1 worldwide
15 sec
max duration · 1080p
12 refs
multimodal inputs
What is Seedance 2.0 and why it matters
Seedance 2.0 is ByteDance's video generation model — the company behind TikTok — launched on February 10, 2026. It's not an incremental update: it uses a Dual-Branch Diffusion Transformer architecture that generates audio and video simultaneously and natively, something no competitor in the same category did before. It accepts up to 12 reference files in a single generation: 9 images, 3 videos, and 3 audios.
Analysts called it the “DeepSeek moment for AI video”: just as DeepSeek shook up the LLM market, Seedance 2.0 triggered drops in U.S. tech stocks and gains in Chinese exchanges. In the Artificial Analysis benchmark, it surpassed Google Veo 3, OpenAI Sora 2, and Runway Gen-4.5 in both major categories.
Why it's technically different
Most models generate silent video and add audio in post-production. Seedance 2.0 generates both in the same step, with native lip-sync in 8+ languages. Additionally, its @ system allows referencing characters, movement, camera, and sound separately — director-level control, not prompt-user control.
The scandal timeline
Feb 7, 2026
Limited beta in China. Within 48 hours, clips of Tom Cruise fighting Brad Pitt, Friends characters as otters, and Will Smith battling a spaghetti monster go viral.
Feb 10, 2026
Official launch. Deadpool & Wolverine screenwriter Rhett Reese posts: “I hate to say it. It's probably over for us.” The Motion Picture Association demands ByteDance cease infringing activity.
Feb 13, 2026
Disney sends a cease-and-desist letter accusing ByteDance of “virtual theft of Disney IP.” Paramount Skydance adds accusations over Star Trek, South Park, and Dora the Explorer.
Feb 16, 2026
ByteDance states it “respects intellectual property rights” and announces filter reinforcement. Temporarily restricts inputs with real faces.
Mar 16, 2026
U.S. senators Marsha Blackburn and Peter Welch demand by letter that CEO Liang Rubo immediately shut down Seedance, calling it “the most egregious example of copyright infringement by a ByteDance product to date.”
Mar 15–25, 2026
Global launch pause while engineers and lawyers redesign IP filters. Third-party red-teaming and C2PA standard invisible watermarks are implemented.
Mar 26, 2026
Relaunch on CapCut for Brazil, Indonesia, Mexico, Philippines, Thailand, and Vietnam. Two days after OpenAI shut down its Sora app, leaving the AI video market increasingly dominated by Chinese companies.
Why Hollywood reacted so strongly
The problem wasn't just quality — it was precision. A content creator who uploaded a photo of their office entrance discovered the model generated the other side of the building (which they had never photographed) and replicated their voice exactly. That's not a creative filter: it's identity generated without consent.
“In a short time, a person will be able to sit in front of a computer and create a movie indistinguishable from what Hollywood produces today.”
— Rhett Reese, screenwriter of Deadpool & Wolverine and Zombieland
The question that also arose among industry professionals was technical: were the viral clips pure generation or video-to-video workflows with green screen stuntmen? Seedance showed R2V workflow examples on its site, suggesting some of the visual impact may come from real video references — something relevant for anyone evaluating the model for production.
What the model can do today
Max resolution
1080p (demo up to 2K in Chinese version)
Max duration
15 seconds · 6 aspect ratios
Simultaneous inputs
9 images + 3 videos + 3 audios + text
Native audio
Yes — lip-sync in 8+ languages, including Spanish
Where to access
CapCut, Dreamina, Pippit · API via third parties
Current restrictions
No real face generation · No unauthorized IP · Not available in the U.S. yet
The competitive landscape in April 2026
Seedance 2.0 leads in multimodal control and benchmarks, but the market has four serious options with distinct roles. Here's the practical breakdown for producers:
| Model | Main strength | Best for | Ref. price |
|---|---|---|---|
| Seedance 2.0 #1 Elo | Multimodal control · native audio · narrative consistency | Advertising, musicals, template-based work | Variable · third-party APIs |
| Kling 3.0 | Human movement · videos up to 3 min · low price | Social media · rapid prototyping · volume | ~$7/mo |
| Veo 3.1 | Photorealism · 4K 60fps · color science | Final deliverables · premium production | $250/mo |
| Runway Gen-4.5 | Professional editing tools · post-pro integration | Post-production workflows · filmmakers | $12–76/mo |
The most relevant market data right now: OpenAI shut down its Sora app on March 24, 2026, two days before Seedance's relaunch. The AI video space is increasingly dominated by Chinese companies — ByteDance and Kuaishou — and Google. Western players are falling behind.
“Many production teams already use multiple models: Seedance 2.0 for reference-based and template work, Kling 3.0 for rapid prototyping, and Veo 3.1 for final deliverables.”
— WaveSpeed AI, 2026 video model comparison
Has another model surpassed it?
Not yet. As of April 2026, Seedance 2.0 holds the highest Elo in both categories according to Artificial Analysis, above Veo 3, Sora 2, and Runway Gen-4.5. Kling 3.0 is the closest competitor in overall quality and surpasses it in price and API availability.
What is clear is that the update cycle in this space is weeks, not months. Seedance's advantage today is no guarantee of anything in 60 days. For anyone working in production, the smart move is not to marry a model — it's to know the full ecosystem and know when to use which.
Takeaway for AI ad production
Seedance 2.0 is the most powerful model today for reference-based work, character consistency, and camera control. But its limited availability — no U.S., API only via third parties — makes it less predictable for commercial production workflows that need stability. Kling 3.0 remains the most practical option for volume. Veo 3.1 for maximum delivery quality.