ModelTC /LightX2V

About LightX2V

LightX2V handles open image and video models on your own GPU. It covers text-to-video, image-to-video, text-to-image and image editing. You download the model weights yourself and need a suitable GPU. It's built for engineers, not for cutting a drama timeline.

Best for: Engineers and researchers who run or tune open image and video models on their own GPUs.

• Runs text-to-video, image-to-video, text-to-image and image editing
• Adds day-0 support for models like MiniMax-H3, LTX-2 and Qwen-Image
• Cuts steps with distilled LoRAs (small add-on models) for 4 or 8 steps
• Shrinks models with FP8, NVFP4 and GGUF quantization
• Splits work across GPUs with parallelism and block-level offload
• Runs on NVIDIA, AMD ROCm, Ascend, Intel and other accelerators

1. 1
   Install from Git with the command below

   pip install -v git+https://github.com/ModelTC/LightX2V.git

   Copy
2. 2
   Or clone and build: the command below, then cd LightX2V and uv pip install -v .

   git clone https://github.com/ModelTC/LightX2V.git

   Copy
3. 3
   Download the model weights from HuggingFace and point the pipeline at the local model path
4. 4
   Run a generation script, such as the MiniMax-H3 4-step 768p distilled LoRA example