Seaweed

Seaweed is a breakthrough foundational model for AI video generation, capable of creating 720p to 2K resolution videos from text, images, or audio inputs. Its 7B-parameter diffusion transformer architecture excels at human-centric scenes—syncing lip movements to audio, maintaining character consistency across shots, and generating 20-second native clips (extendable to 1 minute). Unique features include first/last frame conditioning for controlled transitions and real-time generation at 24fps.

Beyond creative applications, Seaweed enables precise camera trajectory control for virtual world exploration and enhances 3D consistency through CGI-trained physics. Researchers and studios can leverage its multimodal training (video+image+text) for everything from animated storytelling to interactive simulations.

Technical Highlights:
Long-Form Narrative: Maintains continuity across multi-shot sequences

Hybrid Inputs: Text prompts + reference images + audio synchronization