On June 23, 2026, ByteDance’s Volcano Engine held the FORCE conference in Beijing. At the event, ByteDance announced several major AI advances, including the video‑generation model Seedance 2.5, the Doubao large‑model 2.1 series (including a Pro edition) and the image‑generation model SeedDream 5.0. These releases mark rapid iteration by ByteDance in multimodal AI and enterprise‑grade agents, further consolidating its leading position in video generation, general‑purpose agents and content‑creation toolchains. Volcano Engine president Tan Dai emphasized at the conference that only when model capabilities cross a “tipping point” can they truly meet production‑scene demands, and this round of releases exemplifies that philosophy.
Seedance 2.5: Video generation enters the long‑sequence era — 30‑second native output is the headline feature
Video generation has long been one of the challenges in multimodal AI. The Seedance series has drawn attention since its launch, and the release of Seedance 2.5 was widely regarded as the conference’s highlight. The model is currently in global enterprise internal testing and is expected to go live in early July, supporting a single native generation of a complete 30‑second video.
Compared with Seedance 2.0 (maximum about 15 seconds), version 2.5 doubles the length. This breakthrough significantly reduces the fragmentation and stitching costs associated with longer video production, improving narrative coherence and visual consistency. In practice, users no longer need to generate multiple segments and splice them in post‑production to obtain a coherent long shot or a complete short story, which is revolutionary for film production, advertising and short‑drama creation.
Key innovations include:
- Greatly enhanced multimodal reference support: Supports importing up to 50 full‑modal reference materials simultaneously (text, images, audio, video, etc.). This greatly improves controllability and refinement in generation; users can guide style, character consistency and motion details with abundant references, reducing randomness and enabling user‑led precise creation.
- 3D white‑model support and improved controllability: Adds 3D white‑model support and optimizes complex interactions, motion generation and instruction following. Character consistency, style uniformity and multi‑shot storytelling perform better, meeting industrial delivery standards.
- Resolution and quality upgrades: Builds on Seedance 2.0’s native 4K output with improvements in detail (such as hair strands and clothing texture), motion smoothness and color rendition to approach professional equipment quality; supports 10‑bit output for color grading flexibility.
- Expanded application ecosystem: Works with a newly launched AI rights platform to support authorized IP remixes. For example, templates based on Stephen Chow’s classic film IP have been made available on Douyin and Jianying, with daily creation counts already exceeding 100,000, demonstrating a closed loop from tools to ecosystem.
These innovations not only push AI video from “short demo” toward “industrial production” but also lower the creation threshold. Bulk e‑commerce content generation, marketing asset production and UGC creation will see efficiency gains. According to the company, the Seedance series leads the industry in motion stability, audio–video joint generation and director‑level control (such as lighting and camera motion).
Doubao 2.1 Pro: Crossing the production‑grade tipping point, targeting Claude Opus 4.6
On the general‑model front, Doubao 2.1 Pro was another focal point of the release. Volcano Engine stated that the model achieves capability leaps in three core areas — coding, agents and VLM (vision‑language models) — and that on multiple benchmarks it performs better than or close to Claude Opus 4.6, already meeting conditions for full production deployment.
Key technical innovations and capability improvements:
- Coding and engineering delivery: Stronger requirement understanding, long‑term planning and code‑engineering capabilities, supporting end‑to‑end delivery of complex projects. Combined with a dedicated code model, it covers diverse scenario needs.
- Agent capabilities: Enhanced end‑to‑end task handling with support for long‑context understanding. For example, it can watch a two‑hour video and perform end‑to‑end editing, demonstrating powerful multimodal processing — laying a solid foundation for enterprise‑office and content‑creation AI agents.
- VLM and multimodality: Visual understanding and generation abilities are significantly enhanced, forming synergy with Seedance and SeedDream.
- Practical deployment: APIs are open and integrated into the Coze ecosystem. Volcano Engine also launched a professional Doubao edition and office products for enterprise customers. Tan Dai noted that Doubao’s basic daily functions for users (such as search Q&A) will remain free to further promote AI accessibility.
The release of Doubao 2.1 Pro reflects ByteDance’s AI strategy tilting from consumer toward enterprise. Daily token usage has exceeded 180 trillion, a tenfold year‑on‑year increase, and its MaaS market share leads in China. These figures underline the model’s practical deployment and user recognition.
SeedDream 5.0: Image‑generation upgrade forming a multimodal closed loop
The new image model SeedDream 5.0 (including Pro/Lite versions) debuted alongside the others. It deeply collaborates with the Seedance and Doubao series to help users achieve a one‑stop “image‑to‑video” creation loop.
Highlights:
- Real‑time retrieval enhancement: Introduces real‑time web‑connected retrieval to obtain the latest knowledge and information, accurately meeting time‑sensitive creative needs.
- Improved understanding and generation: Better interpretation of short or ambiguous inputs, improved subject consistency, image–text alignment and multilingual/world‑knowledge capabilities. Supports interactive precise editing and editable layered design generation.
- Professional scenario adaptation: Suitable for professional infographics, architectural design, UI assistance and commercial‑brand marketing, with generation quality approaching industrial standards.
This closed loop means creators can first generate high‑quality image references with SeedDream and then use Seedance to produce videos, greatly reducing cross‑modal switching costs.
Broader ecosystem and strategic significance
These releases are not isolated product updates but a concentrated expression of Volcano Engine’s AI cloud‑native and agent strategy. The conference, themed on “agent‑driven enterprise restructuring,” upgraded the cloud service system for agents, including the Coze platform, TRAE (The Real AI Engineer) and ArkClaw Enterprise Edition.
The ByteDance Seed team, established in 2023, focuses on exploring upper limits of general intelligence across LLMs, vision and world models. Volcano Engine provides one‑stop services from models to deployment to accelerate enterprise AI transformation. Combined with a trustworthy security system (three‑layer protection) and developer incentive programs, these outcomes are accelerating the shift of AI from “usable” to “well‑used.”
Challenges and outlook: Although native 30‑second video output is a leading achievement, longer durations, higher consistency and real‑time interaction remain industry goals. Cost control (for example, Seedance 2.0 Mini reduced per‑second cost by about 50%) and copyright compliance are also key. With APIs fully opening and expanded ecosystem partnerships, ByteDance is likely to capture a larger share of the global AI content‑generation market.
The Volcano Engine 2026 releases not only showcase technical strength but also highlight the innovation vitality of Chinese AI companies in multimodality and enterprise applications. Seedance 2.5’s 30‑second breakthrough, Doubao 2.1 Pro’s production‑grade capabilities and SeedDream 5.0’s collaborative closed loop are poised to spark a new efficiency revolution in the content industry. The industry awaits further real‑world performance and broader commercialization of these models.
▶️ Enjoyed this article? Creating great content takes effort!
Subscribe to our YouTube channel:
AI Rosetta
Your likes, subscriptions and shares mean the world to us! 🙏
For more AI news, visit:
https://www.airosetta.com