Open-Sora
Open-source text-to-video generation model
28.9k|stable|v1.3
Multiple generation modes — Text-to-video, image-to-video, video-to-video, and infinite time generation.
Flexible video specs — Generate 2s to 15s videos from 144p to 720p in any aspect ratio.
Efficient training — Produces 512×512 videos in 3 days with 46-50% cost reduction.
Full pipeline — Data preprocessing, training acceleration, and inference included.
Open-Sora is a fully open-source video generation framework with training code and model checkpoints available. The 2.0 release (11B) matches performance of larger proprietary models while being trainable with just $200K in compute. Includes Gradio UI, ColossalAI acceleration, and support for variable resolution and duration.
GPU required for training and inference; see ColossalAI documentation for detailed requirements
Source on GitHub
pavilion