Artificial Intelligence March 28, 2026

ByteDance Launches Dreamina Seedance 2.0 AI Video Model in CapCut

On March 26, 2026, ByteDance officially rolled out Dreamina Seedance 2.0 – its most advanced AI video generation model – directly inside CapCut, the company’s popular video editing platform. The integration marks a significant escalation in the AI video arms race, arriving just as OpenAI shuttered its competing Sora app. Seedance 2.0 allows creators to generate video clips up to 15 seconds long from text prompts, images, audio clips, and reference videos, complete with synchronized sound, realistic textures, and consistent character rendering across scenes.

What makes this launch particularly notable is its scope. Rather than existing as a standalone tool, Seedance 2.0 is woven into CapCut’s editing features, its new timeline-free Video Studio workspace, the Dreamina AI generation platform, and ByteDance’s marketing platform Pippit. For creators, this means an end-to-end workflow – from ideation to final export – without ever leaving the ecosystem. The model supports six aspect ratios, handles multimodal inputs including up to 9 images, 3 videos, and 3 audio clips per project, and produces output that leads industry benchmarks in motion stability and immersive realism.

The rollout is phased, however, and deliberately limited. CapCut users in Brazil, Indonesia, Malaysia, Mexico, the Philippines, Thailand, and Vietnam get access first. The United States and other major markets are notably absent, a consequence of a recently reported global pause tied to intellectual property concerns and criticism from Hollywood over alleged copyright infringement. In China, the model is accessible through ByteDance’s Jianying app.

What Dreamina Seedance 2.0 Actually Does

At its core, Seedance 2.0 is a unified multimodal audio-video joint generation architecture. That technical description translates into something genuinely practical: you can feed the model a short text prompt – even just a few words describing a scene – and it will produce a video clip with realistic motion, lighting, parallax, and textures. No reference images are required, though adding them dramatically improves output quality and consistency.

The model excels in areas where previous AI video generators have struggled. Motion-heavy content like cooking recipes, fitness tutorials, and action sequences now render with believable physics and timing. Character consistency is maintained across scenes, meaning a brand mascot or recurring figure retains its identity, features, and style throughout a multi-clip project. Lip-synced dialogue and ambient audio generation work in English, Chinese, and Cantonese, with the AI automatically matching sound to visual content.

ByteDance’s own SeedVideoBench-2.0 evaluations place the model in a leading position across text-to-video, image-to-video, and multimodal task categories. While independent third-party benchmarks remain scarce, the breadth of supported inputs and the precision of style replication set it apart from simpler text-to-video tools on the market.

How to Generate Videos with Seedance 2.0

Access is available through dreamina.capcut.com/tools/ai-video-generator, the CapCut app’s AI Video features, or the new Video Studio workspace on CapCut Web. Here’s the step-by-step process:

Launch and prepare your input (under 30 seconds): Open a new project in CapCut or Dreamina. Upload a high-resolution image (1024×1024 pixels or higher recommended for best results), enter a text prompt of 20-50 words for precision, or add a reference video up to 10 seconds long for audio syncing.
Select model and configure settings (10-20 seconds): Choose Dreamina Seedance 2.0 from the model dropdown. Avoid Seedance 1.0 mini if you want full quality – it’s limited to 5-10 second clips. Set your duration to 5, 10, or 15 seconds. Pick from six aspect ratios, including 16:9 for widescreen or 9:16 for TikTok and Reels. Enable audio sync if you want lip-synced dialogue or ambient sound generation.
Generate (1-5 minutes): Click Generate and wait. Preview the output for realistic parallax, lighting, and motion. You can regenerate up to 3 times if needed. Free tier limits are approximately 5-10 daily generations.
Edit and export (1-2 minutes): Refine the output within CapCut using its editing tools for enhancements or corrections. Download with the invisible watermark intact.

For multi-image storytelling using the Seedance 1.0 fallback, you can upload up to 10 images and the AI will blend them into seamless sequences with prompt-driven transitions.

CapCut’s Video Studio: A Timeline-Free Workspace

Alongside the Seedance 2.0 integration, CapCut introduced Video Studio – a canvas-based production workspace that eliminates the traditional timeline editing paradigm. This is not a minor UI tweak. Video Studio is designed as a complete creative environment where every stage of production happens on a single unlimited canvas.

The workspace includes an AI agent for ideation, writing, and story structuring; a built-in storyboard feature for shaping plot; industry-leading image and video generation models with omni-reference support; and a full editing toolkit for frame-level refinement. The idea is that creators can move from concept to export without switching tools or mental models. CapCut is offering free credits to get started, with tiered pricing beyond the initial allocation.

Safety Controls and IP Restrictions

ByteDance has implemented several safety measures that directly shape how the model can be used – and where it’s available.

Real face blocking: The model will not generate videos from images or videos containing real human faces, a restriction aimed at preventing deepfakes and unauthorized likenesses.
IP protection: CapCut blocks unauthorized generation of intellectual property. The limited initial rollout markets suggest these protections are still being refined, particularly for the U.S. market where copyright scrutiny is most intense.
Invisible watermarking: All content produced by Seedance 2.0 carries an invisible watermark that persists when shared off-platform, enabling identification and supporting takedown requests from rights holders.

The fact that the U.S. remains excluded from the initial rollout is telling. A reported global pause preceded this launch, driven by Hollywood criticism over alleged copyright infringement. ByteDance has committed to partnering with experts and creative communities to iterate on the model’s capabilities and restrictions as the rollout expands.

How Seedance 2.0 Compares to the Competition

The timing of this launch is impossible to ignore. OpenAI’s Sora app – once positioned as the flagship AI video generator – has been shut down. That vacuum creates an opening ByteDance is clearly aiming to fill, but the competitive landscape extends beyond just Sora.

Feature	Dreamina Seedance 2.0	OpenAI Sora (Pre-Shutdown)	Generic AI Video Generators
Supported Inputs	Text, images (up to 9), videos (up to 3), audio (up to 3)	Primarily text-to-video	Basic text or single image
Max Clip Duration	15 seconds	Not officially specified at shutdown	Varies, typically 5-10 seconds
Character Consistency	Maintained across scenes via identity locking	Limited continuity between generations	Minimal to none
Audio Sync	Lip-sync, ambient sound, emotional matching	Not a core feature	Rarely supported
Editing Integration	End-to-end in CapCut Video Studio	Standalone app (now discontinued)	Requires external editing tools
Safety Controls	Face blocks, IP restrictions, invisible watermarks	Less emphasis pre-shutdown	Variable, often unrestricted
Availability	Phased: SE Asia and Latin America first	Discontinued	Broadly available

The key differentiator is controllability. Where generic tools produce one-off clips with limited consistency, Seedance 2.0 is designed for serialized, professional content where characters, styles, and scenes need to hold together across multiple generations.

Prompt Engineering and Best Practices

Getting strong results from Seedance 2.0 requires more than typing “cool video” into a text box. Prompt quality directly determines output quality, and there are specific strategies that yield consistently better results.

The recommended prompt structure follows a roughly 70/20/10 ratio: 70% descriptive action and scene detail, 20% style direction (such as “cinematic” or “anime”), and 10% audio cues. A vague prompt like “fast car” produces generic, unusable output. A prompt like “silver Porsche 911 drifting on wet asphalt at dusk, dynamic camera pan, tire screech audio” gives the model enough specificity to produce something genuinely compelling.

Image quality matters enormously. Blurry images below 512×512 pixels cause visible artifacts. Always start with high-resolution seeds above 1024×1024 pixels.
Don’t skip audio sync. Enabling it early lets the AI handle lip-sync and emotional sound matching automatically, saving hours of manual editing.
Respect the 15-second limit. Attempting to exceed it will fail. Instead, generate multiple 15-second clips and stitch them together in CapCut’s editor.
Iterate deliberately. Plan on 2-3 regeneration attempts per clip. Test ideas from sketches or rough concepts before committing to a full production workflow – this alone can save 50-80% of production time compared to filming first.

Real-World Applications Taking Shape

Early adopters are already finding practical applications that go beyond novelty. Creators are using Seedance 2.0 to prototype video concepts from sketches before committing to expensive shoots – testing camera angles, lighting, and pacing in AI-generated drafts. Marketing teams are leveraging the Pippit platform integration to produce product overview videos that replicate complex visual effects or anime-style aesthetics from reference clips, all without advanced editing skills.

Serialized storytelling is another strong use case. The model’s ability to lock in character identity, features, and style from reference inputs means a recurring character can navigate from cityscapes to forests across multiple clips while remaining visually consistent. In China, early access VIP users on the Jianying app are generating lip-synced narration for dialogue and even singing sequences, with the AI matching facial micro-expressions to custom voice samples.

What Comes Next

The phased rollout strategy signals that ByteDance is playing a longer game. Expanding to additional markets depends on resolving the IP compliance issues that prompted the initial global pause. Full global access is expected after further safety and copyright tweaks, though no specific timeline has been announced.

The broader trend is clear: AI video generation is moving from standalone experimental tools into deeply integrated editing ecosystems. CapCut’s approach – embedding Seedance 2.0 into its editor, Video Studio, Dreamina platform, and Pippit marketing tools – creates a unified workflow that standalone generators cannot match. The model’s leadership in SeedVideoBench-2.0 evaluations across multiple task categories suggests ByteDance is investing heavily in maintaining that technical edge.

For creators, the practical takeaway is straightforward. Seedance 2.0 represents the most accessible entry point to professional-grade AI video generation currently available, with the caveat that geographic availability remains limited and independent benchmarks are still needed to verify ByteDance’s performance claims. The invisible watermarking and face-blocking restrictions add meaningful safety guardrails, but the absence from the U.S. market underscores that the technology’s relationship with intellectual property law remains unresolved. This is a tool worth watching closely – and for those in supported markets, worth testing immediately.