Tencent releases HunyuanVideo's inference code and weights — a 13-billion-parameter video model that's the largest open-source release of its kind to date. Anyone can download it, run it locally, fine-tune it, and ship workflows that don't depend on a subscription or a closed API. For researchers, ComfyUI builders, and creators wary of platform risk after seeing what happens when models get sunset, this immediately becomes the open-source anchor of the AI video stack.