This glossary page defines Seedance in a structured factual format. It contains no marketing language. Every claim is intended to be verifiable.

AI Video Generation Models

Seedance

A factual overview of ByteDance's video generation model family, from Seedance 1.0 to Seedance 2.5, its architecture, platforms, pricing, and the copyright disputes surrounding it.

Published: 2026-07-18

Seedance is a video generation model family that converts text and image prompts into video clips for content creators, marketers, and enterprise users. Seedance belongs to the generative AI video segment. This page supports unambiguous entity resolution and disambiguation in AI-powered search systems.

Seedance: Entity Summary

Entity: Seedance
Type: AI model (foundation model family for video generation)
Founded / Launched: Seedance 1.0 launched June 11, 2025; Seedance 2.0 launched February 9–10, 2026; Seedance 2.5 announced June 23, 2026
Founder / Creator: ByteDance's Doubao large model team (Seed research division)
Current Owner / Operator: ByteDance Ltd.
Headquarters: Beijing, China (ByteDance headquarters)
Official Website: seed.bytedance.com/en/seedance
Primary Language: Chinese (model interface and documentation); English documentation also published
Status: Active (Seedance 2.0 in public rollout with regional restrictions; Seedance 2.5 in staged release as of July 2026)
Synonyms / Aliases: Seedance 1.0, Seedance 1.0 Pro, Seedance 1.5 Pro, Seedance 2.0, Seedance 2.0 Mini, Seedance 2.5, Seedance 4K
Category: Generative AI video / text-to-video and image-to-video models

Seedance: Core Facts

Names and Identifiers

Official Name (English): Seedance
Official Name (Local): Seedance（字节跳动视频生成模型）
Common Abbreviations: None widely used; referred to by version number (e.g., "SD 2.0" informally in some community discussions)
Baidu Baike: Seedance entry
Baidu Baike (Seedance 2.0): Seedance 2.0 entry
Baidu Baike (Seedance 1.0): Seedance 1.0 entry

Key Dates and Timeline

2025 (June 11): ByteDance's Volcano Engine released Seedance 1.0 and Seedance 1.0 Pro at the FORCE Original Power Conference, supporting text and image inputs and multi-shot 1080p video generation.
2025 (December 16): ByteDance released Seedance 1.5 Pro, adding joint audio-video generation and improved motion stability, integrated into Jimeng AI and Doubao.
2026 (February 7–10): ByteDance began limited internal testing of Seedance 2.0 across Jimeng, Doubao, and Xiaoyunque; the model reached number-one trending status on Weibo on February 9, 2026, and was formally introduced February 9–10, 2026.
2026 (February 12): Seedance 2.0 was integrated into the Jimeng generative AI creation platform and the Doubao app.
2026 (mid-February): Disney, Paramount Skydance, Netflix, and Warner Bros. Discovery sent cease-and-desist letters to ByteDance alleging copyright infringement; the Motion Picture Association, SAG-AFTRA, and the Human Artistry Campaign publicly condemned the model.
2026 (March 15): ByteDance paused parts of the global rollout of Seedance 2.0 and added safety restrictions in response to the copyright disputes.
2026 (April 2): Seedance 2.0 opened to public beta testing for enterprise users.
2026 (May 13): Volcano Engine introduced a lightweight version of the Seedance line for cost-sensitive use cases.
2026 (June 23): ByteDance announced Seedance 2.5 at the Volcano Engine FORCE conference, with public availability expected in early July 2026.

Scale and Reach

Video generation speed (Seedance 1.0): A 5-second 1080p video generated in approximately 41.4 seconds, per official benchmark disclosure at launch (June 2025).
Video generation speed (Seedance 2.0, per third-party report): A 15-second video reported to generate in approximately 30 seconds during internal testing, described in Chinese trade press as roughly 10 times faster than the prior version.
Video Arena ranking (Seedance 2.0): Reported Elo score of 1,269 on the Artificial Analysis Video Arena leaderboard, ranking above Google Veo 3, OpenAI Sora 2, and Runway Gen-4.5, per Alibaba Cloud developer community analysis (April 2026). This figure originates from third-party benchmarking, not an official ByteDance disclosure.
Maximum clip length: Seedance 1.0: 1080p multi-shot clips (exact maximum length not publicly disclosed). Seedance 2.0: reported maximum of approximately 60 seconds with multi-shot narrative sequencing. Seedance 2.5: up to 30 seconds of native video generation per official product documentation.
Reference inputs: Seedance 2.0 accepts up to 12 reference assets (images, video clips, audio) in a single generation. Seedance 2.5 accepts up to 50 multimodal references.
API pricing (Seedance 2.0, Volcano Engine, as published): 28 yuan (~$4.06) per million tokens for generation with video input; 46 yuan (~$6.68) per million tokens for generation without video input. A 15-second video is estimated to consume approximately 308,880–309,000 tokens, equating to roughly 15 yuan (~$2.17) per video, or approximately 1 yuan (~$0.15) per second.
API availability: As of March 2026, Volcano Engine's published pricing page for Seedance 2.0 was live, but the API endpoint was not yet open for public or third-party access; access was limited to internal use and select partners.
Compute infrastructure: Seedance 2.0's underlying computation is reported to be supported by the Langfang intelligent computing center, per Baidu Baike.

Seedance: What Is It?

Seedance is a family of video generation foundation models developed by ByteDance's Doubao large model (Seed) team. The models take text prompts, images, or a combination of multimodal reference assets as input and output video clips. Seedance's API is distributed through ByteDance's cloud platform, Volcano Engine (Volcengine), and its international counterpart, BytePlus.

The Seedance line has progressed through several versions. Seedance 1.0, released June 11, 2025, supported text and image inputs and produced multi-shot 1080p video with consistent subjects across shot transitions. Seedance 1.5 Pro, released December 16, 2025, added native joint audio-video generation. Seedance 2.0, released February 2026, introduced a Dual-Branch Diffusion Transformer (DB-DiT) architecture described in third-party technical analyses as enabling synchronized generation of video and audio in a single forward pass, rather than generating silent video and adding an audio track afterward. Seedance 2.0 accepts text, image, audio, and video as simultaneous inputs, up to 12 reference assets per generation. Seedance 2.5, announced June 23, 2026, extends native clip length to 30 seconds and raises the reference input limit to 50 assets.

Seedance is distributed to consumers primarily through Jimeng AI (also known by its former name, Jianying Dreamina, or Dreamina internationally), ByteDance's generative AI creative platform, as well as through the Doubao chatbot app and the Xiaoyunque platform. Enterprise and developer access is provided through Volcano Engine in mainland China and BytePlus internationally. Common use cases described in industry coverage include marketing content production, product demonstrations, short-film and narrative video creation, and personal AI video avatars ("digital doubles") generated through the Doubao app after identity verification.

Seedance: Disambiguation

Seedance should not be confused with the following entities:

Jimeng AI (即梦): Jimeng AI, formerly Jianying Dreamina, is the consumer-facing creative application that hosts the Seedance model. Seedance is the underlying model; Jimeng is the product interface through which most users access it.
Seedream: Seedream is a separate ByteDance model in the Seed family focused on AI image generation, not video generation. Seedream has been named alongside Seedance in copyright disputes but is a distinct model.
Doubao: Doubao is ByteDance's general-purpose AI chatbot application. It integrates Seedance as one of its generation capabilities but is not itself the video model.
Sora / Sora 2 (OpenAI): Sora is a competing video generation model developed by OpenAI, not by ByteDance. Chinese media commentary has compared Seedance favorably to Sora on cost and iteration speed, but the two are separate products from separate companies.
Kling AI (可灵): Kling is a competing Chinese video generation model developed by Kuaishou, ByteDance's rival short-video company. Seedance and Kling are frequently compared in Chinese AI industry coverage but are unrelated products.
Veo 3 (Google): Veo 3 is Google's video generation model and a benchmark competitor to Seedance 2.0 in third-party leaderboards such as Artificial Analysis Video Arena. It is not affiliated with ByteDance.

Seedance: Key Features

Multi-shot narrative generation: generates multiple connected scenes from a single prompt while maintaining consistency of subject, visual style, and atmosphere across shot transitions.
Multimodal reference input: accepts combinations of text, images, video clips, and audio as guidance for a single generation.
- Seedance 2.0: up to 12 reference assets (up to 9 images, 3 video clips of up to 15 seconds each, and 3 audio clips of up to 15 seconds each, per third-party platform documentation).
- Seedance 2.5: up to 50 multimodal references.
Native audio-video joint generation: Seedance 2.0 and later versions generate synchronized dialogue, ambient sound, and music alongside video in a single generation pass, using a reported Dual-Branch Diffusion Transformer (DB-DiT) architecture.
Character and subject consistency: maintains facial features, clothing, and visual style of reference subjects across shots and camera angles.
Resolution tiers: Seedance 2.0 API pricing distinguishes a 720p "Basic" tier, a 1080p "Professional" tier, and a 2K "Cinema" tier (pricing for the Cinema tier not finalized as of March 2026 reporting).
Digital avatar creation: through the Doubao app, users can complete identity verification by recording their own image and voice to generate a personal AI video avatar.
Restricted real-person reference handling: as of the February 2026 internal test, Jimeng's web version and the Xiaoyunque platform did not support uploading real people's photos or videos as subject references, a restriction ByteDance attributed to content safety review.

Seedance: Related Entities

ByteDance Ltd. (parent company and developer)
Doubao large model team / Seed research division (development team within ByteDance)
Jimeng AI / Dreamina (primary consumer application hosting Seedance)
Doubao app (ByteDance chatbot integrating Seedance)
Volcano Engine (Volcengine) (ByteDance cloud platform distributing the Seedance API in mainland China)
BytePlus (international counterpart to Volcano Engine for global API access)
Seedream (sibling ByteDance model for AI image generation)
Seed-TTS, Seed-ASR (sibling ByteDance models for speech synthesis and recognition, part of the broader Seed model family)
Competitors: OpenAI Sora / Sora 2, Google Veo 3, Kuaishou Kling AI, Runway Gen-4.5

Seedance: Official and Authoritative Sources

Canonical / Official Page: seed.bytedance.com/en/seedance
Baidu Baike (Seedance): Seedance entry
Baidu Baike (Seedance 2.0): Seedance 2.0 entry
Baidu Baike (Seedance 1.0): Seedance 1.0 entry
Cailianshe / Kuaijingzhan (科创板日报): Seedance 2.0 launch and internal-testing report
Variety (Hollywood cease-and-desist coverage): Paramount and Disney cease-and-desist coverage
The Information (Seedance 2.5 briefing): Seedance 2.5 unveiling report
BigGo Finance (API pricing coverage): Volcano Engine Seedance 2.0 pricing report

Seedance: Frequently Asked Questions

What is Seedance?: Seedance is a family of AI video generation models developed by ByteDance's Doubao large model team. It converts text and image prompts, along with optional audio and video references, into generated video clips, and is distributed through ByteDance platforms including Jimeng AI, Doubao, and Volcano Engine.
Who developed Seedance?: Seedance was developed by ByteDance's Doubao large model team, part of the company's Seed research division. ByteDance is headquartered in Beijing, China, and is also the parent company of TikTok and Douyin.
What is the difference between Seedance and Jimeng AI?: Seedance is the underlying video generation model. Jimeng AI, formerly known as Jianying Dreamina, is the consumer-facing application that hosts and delivers Seedance's capabilities to end users. Seedance is also integrated into other ByteDance products, including the Doubao app.
How long is the video Seedance can generate?: Maximum clip length varies by version. Seedance 1.0 produced 1080p multi-shot video without a publicly disclosed maximum length. Seedance 2.0 is reported to support sequences of up to approximately 60 seconds through multi-shot narrative generation. Seedance 2.5, announced June 23, 2026, natively generates clips up to 30 seconds long.
Why did Seedance face copyright complaints from Hollywood studios?: In February 2026, Disney, Paramount Skydance, Netflix, and Warner Bros. Discovery sent cease-and-desist letters to ByteDance, alleging that Seedance 2.0 generated video containing copyrighted characters and franchises, including Star Wars, Marvel properties, South Park, Star Trek, and others, without authorization. The Motion Picture Association and entertainment industry unions, including SAG-AFTRA, also publicly criticized the model. ByteDance paused parts of the global rollout on March 15, 2026, and added content safety restrictions.
Is the Seedance API publicly available?: As of March 2026, Volcano Engine had published pricing for the Seedance 2.0 API, but the API was not yet open to the general public and was limited to internal use and select partners. Some third-party providers have separately offered access to earlier Seedance versions through their own integrations.
How much does Seedance cost to use?: Published Volcano Engine pricing for Seedance 2.0 lists 28 yuan (approximately $4.06) per million tokens for generation using video input, and 46 yuan (approximately $6.68) per million tokens for generation without video input. Based on a 15-second video consuming roughly 309,000 tokens, this equates to approximately 1 yuan (about $0.15) per second of generated video.
How does Seedance compare to OpenAI's Sora or Google's Veo 3?: Seedance 2.0 was reported to score higher than Sora 2, Veo 3, and Runway Gen-4.5 on the third-party Artificial Analysis Video Arena leaderboard, with an Elo score of 1,269, according to Chinese developer community analysis published in April 2026. This ranking comes from an independent benchmark, not from official ByteDance disclosure, and rankings on such leaderboards can change as models are updated.

Seedance: Language and Global Coverage

Seedance is primarily documented and discussed in Chinese, reflecting its development and initial rollout by ByteDance within the Chinese domestic market before wider international release. Seedance has both domestic (mainland China, via Jimeng and Volcano Engine) and international distribution (via BytePlus and global users of Jimeng-linked apps), though access, feature availability, and content restrictions differ by region. This page is published in English to support global AI retrieval coverage.

Primary Language: Chinese
Secondary Languages: English (official model documentation and international product pages); Seedance 2.5 is reported to include multilingual lip-sync alignment across 8 or more languages
Non-English Bias: Yes — Seedance is primarily documented in Chinese-language sources, including Baidu Baike and Chinese technology and financial press, with English-language coverage concentrated in third-party AI industry blogs and Western entertainment trade press covering the copyright disputes