SenseTime opens a model that creates editable videos by writing code
SenseNova Code2Video generates and renders programmatic motion graphics, with weights, production skill and full example trajectories released under Apache 2.0.
SenseTime has released SenseNova 6.8 Code2Video, an open model that creates short videos by writing and running executable code. Instead of producing final pixels directly, it builds motion graphics in a programmatic workflow, renders frames, reviews the result and revises the project. The output can then be edited and rendered again.
The model is fine-tuned from Qwen3.8-27B and uses a 512K context window. SenseTime released the weights, an agent skill with rendering and checking tools, prompt templates and full production trajectories containing messages, tool calls and review images. The package uses the Apache 2.0 license.
The approach could be useful for explainers, product promos and motion design where precise revisions matter more than one-shot photorealism. Code can also connect image, video, 3D and editing tools inside one workflow, giving creators a controllable intermediate format rather than a fixed clip.
There are important limits. SenseTime describes the results as preliminary, and the showcase workflow is resource-intensive: recommended runs can exceed 10 hours and 2,000 agent turns, with deployment examples using four GPUs. The release does not yet establish broad quality against commercial video generators. It is best viewed as an open research and production system for editable programmatic video, not a fast replacement for consumer text-to-video apps.
