SenseTime open-sources SenseNova 6.8 Code2Video for code-driven video generation

On November 3, 2026, at 23:27, a post appeared on V2EX titled 'SenseNova 6.8 Code2Video Open-Sourced by SenseTime: Generate Video with Code'. According to that post, SenseNova 6.8 Code2Video is an open-source release from SenseTime. The model has 27B parameters and is designed specifically to generate video by writing and modifying code rather than predicting pixels frame by frame. The release reportedly includes full model weights and a companion creation Skill, together with a project page at https://github.com/OpenSenseNova/SenseNova-6.8-Code2Video. The post states that generated results can be revised through multiple rounds of natural language instructions, and that the model can even reverse-engineer a reference video that has no source code at all into an editable code project. It also describes the architecture as a 27B specialist model with 512K context for code generation, aimed at science explainers, motion graphics, and typographic animation. The post opens by referencing community discussion about the high cost of producing a 15-second video via Claude Opus 5.5 code, which reportedly cost close to USD 100, and characterises the new model's results as impressive. These details come from the V2EX post itself and cite no official announcement for verification. As of now, no further coverage of the project has appeared.