Seedance 2.0 API Officially Opens for Public Beta: One Yuan per Second, ByteDance Wants to Turn AI Video into Infrastructure

On April 2, Volcano Engine under ByteDance announced at the “2026 Force Link AI Innovation Roadshow” in Wuhan that the Seedance 2.0 API is officially open for public beta for enterprise users.

This news actually started circulating in various groups on the night of April 1, making many people think it was an April Fool’s joke. But it’s real this time — at 10:00 a.m. on April 2, the application channel opened right on time.

Seedance 2.0 is not new. As early as February 12, it was already integrated into the Doubao app, desktop client, and web version, and ordinary users have been using it for almost two months. But now the API is open, meaning enterprises can integrate it into their own products and workflows. From a “consumer toy” to “enterprise infrastructure,” that’s a big leap.

What Can It Do

First, the capabilities. Seedance 2.0 has four core selling points:

Text-to-video: Input a text description to directly generate a video. People are already familiar with this — since Sora exited the scene, this track has basically been taken over by domestic models.

Image-to-video: Input an image and make it move. E-commerce companies may be most interested — product images directly turned into display videos, eliminating shooting costs.

Original audio-visual synchronization: This is one of Seedance 2.0’s killer features. The generated video comes with matching sound effects and ambient audio—no need for later dubbing. Imagine: you generate a video of waves hitting the shore, and the sound of waves automatically appears.

Character consistency: The same character maintains a consistent appearance across different shots and scenes. Teams making short dramas or animations understand this pain point — previously, with AI-generated videos, the character would “change faces” with each new shot.

Pricing: One Yuan per Second

Now let’s look at what everyone cares about most — the price.

Seedance 2.0 has two pricing tiers:

• Including video input (video editing scenarios): 28 yuan per million tokens
• Excluding video input (pure text-to-video): 46 yuan per million tokens

Pure text-to-video consumes more computing power, so it’s more expensive. Based on the 46 yuan unit price, generating a 15-second video consumes approximately 308,880 tokens, which equals 15 yuan per video.

One yuan per second. Expensive? Depends on the comparison.

Compared to real filming — 15 yuan for a 15-second shot — that’s not even enough for the photographer’s coffee. Compared to other AI video generation services, this price is mid-to-high range, but Seedance 2.0’s visual quality and consistency are certainly first-tier.

Public Beta Rules: Big Gap Between Standard and Advanced Versions

The public beta rules are designed in an interesting way, dividing users into two levels:

Standard Version (Free Application)

• Only available for verified enterprises; individual users not supported
• Default concurrency of 10 (each enterprise can submit up to 10 tasks simultaneously)
• Real human face generation not available
• Custom virtual portrait not available
• Can only use the platform’s public virtual portrait library

Advanced Version (Requires Agreement)

• Higher concurrency limit
• Unlock real human face generation
• Unlock custom virtual portrait creation
• Get official technical support
• Cost: sign a guaranteed cooperation agreement and pay 10% prepayment + 1 million yuan security deposit

A one-million-yuan deposit immediately filters out small teams. But from ByteDance’s perspective, this design makes sense—real human face generation involves portrait rights and deepfake risks, so binding high-fidelity capabilities to institutions with funding and compliance ability acts as insurance.

Volcano Engine President Tan Dai also mentioned in an interview that opening the API externally requires a complete copyright protection system. Based on Doubao’s VLM capability, they built a new copyright protection framework that not only protects a person’s current likeness but also covers versions such as youth and ancient costumes.

The background for this statement: media reports previously noted that due to copyright disputes, ByteDance suspended Seedance 2.0’s overseas release plan. Opening domestically first, with copyright protection in place—this is a lesson learned and a smart move.

Doubao’s Data: Daily Token Usage Exceeds 120 Trillion

At the same event, they also announced that Doubao’s large model now reaches over 120 trillion tokens in daily usage.

How impressive is that? Two years ago, when Doubao first launched, daily consumption was around 120 billion tokens. That’s a 1000-fold increase in two years.

Even more noteworthy is the structural shift—enterprise-level calls are rising rapidly. This means Doubao is not just serving consumer chat users; more and more companies are integrating it into their business processes. The opening of the Seedance 2.0 API will further accelerate this trend.

What It Means for Developers

If you’re building video-related products, the Seedance 2.0 API is worth your attention. Several real-world application scenarios:

E-commerce: Turn product images into presentation videos in one click, mass-produce product short videos. What previously required studios, models, and post-production can now be done with a single API call.

Short dramas/animation: Character consistency + audio-visual synchronization means you can use AI to generate coherent multi-shot narratives. The advanced version with the 1-million-yuan deposit supports real human faces, but the public virtual portrait library is already sufficient for animated-style short dramas.

Marketing: Quickly generate different versions of ad videos for A/B testing. At 15 yuan per video, it’s orders of magnitude cheaper than hiring a full video team.

Education/training: Convert text tutorials into video courses—with audio-visual synchronization, the effect is much better than plain PPT recordings.

OpenAI Hub has already integrated Seedance 2.0. If you want to try it, you can directly call it via the unified OpenAI Hub API:

import requests

response = requests.post(
    "https://api.openai-hub.com/v1/videos",
    headers={"Authorization": "Bearer YOUR_API_KEY"},
    json={
        "model": "seedance-2.0",
        "prompt": "A golden retriever running on a beach at sunset, slow motion, cinematic quality",
        "duration": 15
    }
)

Sora Is Gone, Domestic Models Step In

An interesting industry backdrop: In March this year, OpenAI officially shut down its Sora video model, choosing to focus on ChatGPT and code generation tools. This means the “benchmark” of global AI video generation is temporarily vacant.

Filling that gap are almost entirely Chinese companies. Seedance 2.0, PixVerse V6, Keling, Tongyi Wanxiang — domestic AI video models have collectively surged in the past six months. And not just catching up— in some dimensions (e.g., audio-visual synchronization, character consistency), they’ve actually reached global leadership.

By opening the API this time, Seedance 2.0 is turning “capability showcase” into “commercial infrastructure.” For the entire industry, it’s a clear signal: AI video generation is moving from the “looks cool” demo stage to the “can actually make money” industrial stage.

---

References:

• Sci-Tech Innovation Board Daily - Seedance 2.0 Public Beta for Enterprise Users — Initial report with pricing and basic info
• Blue Whale News - ByteDance Seedance 2.0 API Launches Public Beta, Emphasizes Copyright Protection — Volcano Engine president Tan Dai’s statement on copyright protection system
• The Paper - Seedance 2.0 Finally Opens, But the Conditions Are Beyond Imagination — Detailed explanation of public beta rules, including full comparison of standard and advanced versions
• Tonghuashun - Seedance 2.0 Officially Announces Public Beta: AI Video Says Goodbye to Manual Workshops — Analysis of API opening from an industrialization perspective
• Baidu Baijiahao - Doubao Token Daily Usage Multiplies by 1000x in Two Years — Volcano Engine business data and trillion-yuan revenue goal