Seedance 1.0 debuts as ByteDance’s AI video model, leading benchmarks and outpacing Veo 3

ByteDance, the parent company of TikTok, has introduced Seedance 1.0, a new AI video generation model positioned to compete directly other popular video models. According to benchmark results, Seedance 1.0 ranks first in both text-to-video and image-to-video tasks, outperforming Google's Veo 3, OpenAI’s Sora, and Kuaishou’s Kling 2.0 in prompt accuracy, motion quality, and image sharpness. The model generates multi-scene videos with consistent characters and complex camera movements, and is noted for sticking closely to user instructions.

Seedance 1.0 was trained on a massive dataset of public and licensed video clips, which were filtered to remove logos, subtitles, and violent content. Manual and automated annotations added detail about movement and style to improve output quality. Training was carried out in multiple phases, starting from general video and image data, then fine-tuned for image-to-video tasks, with reinforcement from human feedback selecting better results.

The model can generate 5 seconds of full HD video in about 41 seconds, faster than many competitors, although it does not yet support audio like Veo 3 does. ByteDance plans to integrate Seedance 1.0 into platforms like Doubao and Jimeng, targeting both professionals and general users for tasks such as marketing, video editing, and content creation via voice prompts.