ByteDance launches Seedance 1.0 video generation model

ByteDance’s Seed team launched Seedance 1.0, a foundation model for video generation from text and image prompts, at the company’s Volcano Engine FORCE conference. The model generated 1080p video with native support for multiple shots within a single clip — ByteDance said it could produce two or three shot transitions within a ten-second video while preserving character and scene consistency across the cuts, a capability most competing video generators handled poorly at the time.

ByteDance published benchmark results alongside the launch: on the third-party Artificial Analysis leaderboards, Seedance 1.0 ranked first in both text-to-video and image-to-video evaluation, ahead of rivals including Google’s Veo 3. The company also reported a speed figure — a five-second 1080p clip generated in roughly 41 seconds on Nvidia L20 hardware — positioning the model as competitive on cost and latency as well as output quality, a point ByteDance emphasised given the compute expense typical of diffusion-based video models.

The model reached consumers through ByteDance’s Doubao and Jimeng apps and reached developers through the Volcano Engine API, continuing the company’s practice of shipping frontier models simultaneously into its own products and as a hosted API for outside customers. Seedance 1.0 arrived amid an active contest among Chinese and American labs over video generation, following OpenAI’s Sora and preceding Google’s and Runway’s own iterations later in the year; ByteDance updated the line to Seedance 1.5 Pro in December 2025, adding synchronised audio generation.