SenseTime Open-Sources SenseNova 6.8 Code2Video: Write the Code, Then Render the Video
SenseTime has released SenseNova 6.8 Code2Video, a 27B open model under Apache 2.0 built on Qwen3.8-27B.
Instead of generating pixels, it writes executable code and renders it into short motion-graphics and explainer videos.
Most video models paint frames. SenseTime is taking a different route: have the model write code, then render that code into a video.
SenseNova 6.8 Code2Video is a 27B-parameter open model fine-tuned from Qwen3.8-27B, released under the Apache 2.0 license with weights on Hugging Face. SenseTime calls it the first open-source model for code-driven video creation.
Videos that are written, not drawn
According to SenseTime, the model builds and renders videos by generating executable code, and for now it focuses on short pieces implemented entirely in code: motion graphics, explainers, stylized 2D and 3D scenes and procedural audio. It does not produce photorealistic footage. The company's 90-second launch film, and every clip inside it, was made by the model itself.
Starting from code has a practical upside: every element stays a named object in the source, so changing copy, colors or timing means editing code and re-rendering rather than starting over. SenseTime also notes that code can call image, video, 3D and editing tools, which is where it wants to take the approach.
Who it suits, and who it doesn't
Given those limits, science explainers, charts and course-style animations look like the natural fit. If you want live-action texture, look elsewhere.
The company is candid about the rough edges. It reports qualitative examples only and has no systematic benchmark yet. A full production run takes hours; the reference setup uses four GPUs and runs inside Claude Code with the supplied code-video skill, and a single job can exceed 10 hours and 2,000 turns. SenseTime also says the model trails the strongest proprietary systems on visual polish and long-horizon reliability.
So this is more a direction worth watching than a tool for mass-producing clips today. But a 27B model under Apache 2.0 means small teams and researchers can deploy and adapt the whole write-code-to-make-video workflow themselves, and that alone gives the release real value.