00:00In a fast-evolving world of AI, Alibaba's one project has emerged as a trailblazer in
00:03open-source video generation. It all began with 1.2.1, a model that laid a strong foundation for
00:09creating dynamic videos from text and images. But in July 2025, Alibaba took it to the next
00:15level with 1.2, a revolutionary upgrade that's redefining what's possible. 1.2 introduces a
00:20groundbreaking mixture of experts' architecture, splitting the video creation process into
00:24specialized models, one for rough layouts and another for fine details. This clever design
00:29doubles the model's power to 27 billion parameters while keeping the computational cost low,
00:33making high-quality video generation faster and more efficient. It doesn't stop there.
00:381.2 was trained on a massive dataset, 65% more images and 83% more videos than 1.2.1,
00:44unlocking its ability to handle complex motions from hip-hop dancing to sweeping cinematic shots,
00:48with unmatched realism. The model also brings cinematic flair, letting creators fine-tuned
00:53lighting, composition and colour for professional-grade visuals. The star of the show is 1.2's
00:58dedicated to a Vuminous 5B model, a lightweight powerhouse that generates 700 trend-type videos
01:02at 24 frames per second, even on consumer GPUs like the RTX 4090. With a high-compression design,
01:09it delivers stunning text-to-video and image-to-video results in under 9 minutes,
01:12making it a game-changer for filmmakers, researchers and creators. Fully open-source,
01:17under the Apache 2.0 license, 1.2.2 is accessible on GitHub, Hugging Face and Model Scope,
01:22with seamless integration into tools like ComfyUI and Diffusers. From its innovative architecture to its
01:27community-driven spirit, 1.2.2 is pushing the boundaries of ReiI video, inviting creators
01:31worldwide to shape the future of storytelling.
Comments