GitHub radar
MAGI-2 Preview: Open Video + Audio Generation by Sand.ai
Sand.ai open-sources MAGI-2 Preview, a 114B-parameter model that generates 10-second video clips with synchronized audio from text or image prompts. Weights are on Hugging Face; eight NVIDIA Hopper GPUs required to run locally.
MAGI-2 Preview is a unified audio-video generation model from Sand.ai with 114 billion parameters that activates only 6 billion per token, using a Mixture-of-Experts architecture called MagiMoE. It generates ten-second video clips from a text prompt (T2V) or a text prompt plus a reference image (I2V), with audio produced and mixed in the same pass. Inference runs in two stages: a preview denoising pass followed by an upscale to 1080p. Weights (307 GB total) are published on Hugging Face under Apache 2.0.
Why a vibe-coder should care
If you're building an app that needs to turn text descriptions into short video clips with audio, MAGI-2 Preview is one of the first fully open-source models to handle both in a single architecture. The hardware requirement is high — eight Hopper-class GPUs — so this is realistically a cloud-deployment or research scenario rather than a laptop experiment.
▌ More finds