I recently tested the open-source AI model Wan 2.2, which can generate cinematic video from a single image. I set it up inside ComfyUI using just 8GB VRAM and documented the full process. It includes model installation, workflow setup, and a fix for a missing dependency (sageattention) using Gemini CLI. I also ran a full test using a wolf image + prompt with motion effects.