SenseNova open-sources Code2Video, a 27B model that makes videos by writing code
quietsam8 · reddit · 2026-10-29
SenseNova open-sourced Code2Video, a 27B model that generates videos by writing code for each shot, rendering it, reviewing frames, and revising—rather than generating pixels.
- The reasoning: a ball's bounce comes from physics in the code, not appearance priors; text, formulas, and chart numbers are typed from source so they don't come out as gibberish; every element is a named object, so edits are small changes instead of full regeneration;
- The release includes weights (a fine-tune of Qwen3.8-27B), a skill file guiding an AI agent through the production process, a template turning short requests into full briefs, and full production logs with messages, tool calls, and review images;
- Built for motion graphics, explainers, and stylized 2D/3D scenes. Repo on GitHub.
More from Multimodal
- Few-Shotting a Procedural Lydian Mode Generator from Exported Ableton MIDI — teropa · 2026-10-29
- NUS Proposes Wiener-Informed Decoding to Speed Up Pixel-Space Flow Model Training — NationalUniversityofSingapore · 2026-10-29
- NVIDIA open-sources Nemotron-Labs-Diffusion-Uni, unified diffusion LM for multimodal — siyan_zhao · 2026-10-29
- Seedance 2.5 on Pippit AI used to produce a cinematic brand promo, lowering the bar for video ads — anthara_ai · 2026-10-29
- Recreating a Higgsfield Festival Nominee with MiniMax H3 in ComfyUI — no3us · 2026-10-29
- Suno Launches MCP Beta for In-Workflow Song Generation — suno · 2026-10-29