SenseTime Open-Sources 27B SenseNova-6.8-Code2Video That Programs Each Shot Instead of Generating Pixels

SenseTime released SenseNova-6.8-Code2Video, a 27B image-text-to-text model built on a hybrid Qwen3.5 architecture that generates videos by writing and iteratively revising per-shot code rather than directly producing pixels, then renders and critiques its own frames. The weights shipped to Hugging Face on October 27 with Docker deployment examples for LightLLM, vLLM and SGLang, and community members quickly posted quantized GGUF variants. SenseTime is positioning the model as an open-source alternative to closed video generators for workflows that need precise, controllable motion.