LightX2V

Why we included this project

LightX2V is a practical choice if you want to self-host video generation without wrestling a pile of separate model repos. It wraps a range of diffusion and autoregressive models, including Wan 2.2, LTX, Hunyuan, Qwen-Image, and MiniMax-H3, behind one pipeline that covers text-to-video, image-to-video, text-to-image, and image editing. The efficiency work is where it gets interesting: FP8/NVFP4 quantization and distilled LoRAs cut inference steps substantially, and features like disaggregated deployment, tensor and sequence parallelism, and model offloading help spread multi-GPU workloads. Apache-2.0 licensing, Docker images, and bilingual docs lower the barrier for small teams that prefer running their own stack to paying for hosted APIs.