Local NVFP4 Qwen 27B replicates Opus-style code-rendered video, no API needed

yzjJosh · reddit · 2026-10-07

The "Opus 5.5 makes videos" trick doesn't generate pixels: the model writes a self-contained HTML/Canvas scene where every frame is a pure function of time, a headless browser captures frames, and ffmpeg encodes the MP4. The author ran the same pipeline on a local NVFP4-quantized Qwen 27B — no API, no GPU rental — producing a 2-minute 1080p explainer of how GPS works with deterministic canvas scenes, TTS narration, and WebAudio-synthesized BGM, showing code-rendered video doesn't need a frontier cloud model.

Original post →

More from coding & agent

coding & agent channel →