ByteDance is pushing deeper into AI video generation just as OpenAI pulls back. The company confirmed Thursday that Dreamina Seedance 2.0, its new audio and video model, is now rolling out inside CapCut, according to TechCrunch AI.
The timing is notable. OpenAI recently shut down its Sora app, effectively stepping away from the consumer video generation space. ByteDance is stepping right in.
But there’s a catch: the rollout is limited. Dreamina Seedance 2.0 is launching first in Brazil, Indonesia, Malaysia, Mexico, the Philippines, Thailand, and Vietnam, with more markets coming later. The U.S. is conspicuously absent from that list.
Why the limited launch?
TechCrunch AI reports that a planned global rollout was paused while ByteDance works through intellectual property issues. Hollywood has criticized the model over alleged copyright infringement, which likely explains the careful, market-by-market approach.
In China, the model is already available through ByteDance’s Jianying app.
What Dreamina Seedance 2.0 can do
Here’s what creators get access to:
- Text-to-video generation: describe a scene in a few words, no reference images needed
- Prompt, image, or reference video inputs: multiple ways to draft and edit content
- Realistic rendering: textures, movement, lighting across various angles and perspectives
- Footage enhancement: edit, improve, or correct existing creator footage
- Concept testing: turn early sketches or ideas into video before actual filming
- Action-focused content: cooking, fitness, product demos, and motion-heavy scenes where AI video has historically struggled
At launch, clips max out at 15 seconds across six aspect ratios. Not long-form content by any stretch, but enough for social media creators and quick marketing videos.
Inside CapCut, the model will appear in editing features like AI Video and generation tools like Video Studio. It’s also coming to Dreamina (ByteDance’s AI generation platform) and Pippit (its marketing platform).
Safety guardrails and IP protections
ByteDance is clearly trying to get ahead of the copyright conversation. Several restrictions are built in:
- No real faces: the model won’t generate video from images or videos containing real people’s faces
- IP blocking: CapCut will block unauthorized generation of intellectual property
- Invisible watermarks: all generated content carries a hidden watermark to identify AI-made videos when shared off-platform
As TechCrunch AI points out, if these restrictions were fully working, the model would likely already be available in the U.S. More tweaks are clearly still in progress.
The watermarking is a smart move. It gives rights holders a way to identify and flag AI-generated content for takedown requests if copyrighted material slips through.
Why this matters
ByteDance owns both the creation tool (CapCut) and the distribution platform (TikTok). That’s a vertically integrated pipeline most competitors can’t match. Putting a capable AI video model directly inside CapCut, where millions of creators already edit their content, removes friction that standalone AI video tools face.
The limited geographic rollout and Hollywood pushback show the tension between shipping fast and navigating IP law. ByteDance says it will partner with experts and creative communities as the model expands to iterate on its capabilities.
What stands out here is the strategic contrast: OpenAI retreating from consumer video generation while ByteDance advances. The company with the largest short-form video platform in the world is now embedding AI video creation right where creators already work.
More details are available in the original TechCrunch AI report.