Favicon of SkyReels

SkyReels

AI video generator with a hosted web app for synchronized sound and reference control, plus downloadable models for local GPU inference.

Screenshot of SkyReels website

SkyReels-V2 is a family of downloadable video models for developers and researchers who want to run generation on their own GPUs. It supports text-to-video, image-to-video and extended video generation. Its autoregressive Diffusion Forcing architecture generates longer sequences by continuing video segments.

The official repository provides Python inference code for single-GPU and multi-GPU use, with xDiT USP for acceleration across GPUs. Text-to-video and image-to-video models also work with Hugging Face Diffusers. Weights are available through Hugging Face and ModelScope.

GPU memory requirements depend on model size, output resolution, frame count and offloading. The documentation reports about 14.7GB peak VRAM for a 1.3B model at 540p and roughly 43–51GB for tested 14B workflows. Reducing the base frame count can lower peak memory at a possible quality cost. These figures describe specific tested settings rather than universal minimums.

The repository directs users to the Skywork Community License. It supports commercial use subject to those custom terms; downloadable weights should not be treated as unrestricted Apache or MIT models.

SkyReels also offers a hosted web application with synchronized audio, camera control and image, video and audio references. Those features belong to the remote service. The downloadable V2 code covers text and image inputs for local video generation.

Similar to SkyReels