Seedance is a Text-to-video generation and image-to-video generation model developed by ByteDance, the Chinese technology company that owns TikTok and its domestic counterpart Douyin. Seedance was released in 2025 as part of ByteDance's broader "Seed" family of foundation models, which also includes image-generation and language models developed by the company's Seed research team. Seedance quickly drew attention for placing at or near the top of community-run benchmarks and arena-style leaderboards that rank text-to-video models by human preference, competing directly with Kling, Veo, and Sora.
Capabilities
Seedance generates video from text prompts and from a starting image, producing clips with attention to motion coherence, camera movement, and multi-shot consistency, capabilities that had been persistent weak points for earlier generations of diffusion-based video generators. ByteDance emphasized Seedance's ability to follow complex, multi-element prompts describing several simultaneous actions or camera behaviors, and its relative efficiency, generating video at competitive speed relative to rival models given similar output length and resolution. Later Seedance versions extended supported resolutions and durations and added finer stylistic and camera controls aimed at professional content creators and advertisers, an audience ByteDance had direct commercial interest in serving given its ownership of TikTok and other short-video platforms.
Competitive position
Seedance entered a text-to-video field that had become intensely competitive through 2024 and 2025, with multiple Chinese labs, including Kuaishou's Kling and other entrants, alongside Western labs' Sora and Veo, iterating on overlapping capabilities within months of one another. Independent evaluators and comparison platforms found Seedance frequently ranked among the top one or two video models by human preference through much of 2025, a result that some industry commentators cited as a signal that ByteDance's substantial compute and video-data resources, drawn in part from its scale as a video platform operator, gave it a structural advantage in this specific generative domain. ByteDance made Seedance available both through its own consumer-facing apps and through an API aimed at developers and enterprise customers building video-generation features into other products.
Reception and context
Seedance's strong benchmark performance was cited, alongside DeepSeek's DeepSeek-R1 and Alibaba's Qwen, as further evidence of the rapid rise of Chinese AI labs across multiple modalities, complicating earlier assumptions that the most resource-intensive generative categories, such as video, would remain dominated by the largest well-funded US labs. As with other advanced video-generation systems, Seedance raised concerns familiar to the broader text-to-video field around Deepfake misuse, copyright status of training data, and the difficulty of distinguishing AI-generated video from real footage, concerns ByteDance addressed through content moderation policies and watermarking of generated output, though independent verification of training data provenance remained limited, consistent with industry-wide opacity around video-model training sets.