SkyReels-V1

Why we included this project

SkyReels-V1 targets the human-centric slice of video generation, a space where most open-weight models fall back on generic scenery. It pairs a text-to-video model with an image-to-video model, so you can either write a prompt from nothing or bring a still portrait to life, with an eye toward believable facial expressions and cinematic shots. The release includes model weights, inference code, a Gradio web demo, and a parallel inference setup on Diffusers that works on a single RTX 4090 or stretches across multiple GPUs. For filmmakers, content teams, and developers testing character-driven clips, the mix of focused training data and practical tooling makes it a dependable place to start.