Qwen quantized model on 16GB VRAM recreates Opus-style motion graphics in ~2 hours
Yhojm · reddit · 2026-11-09
A Reddit user ran a quantized Qwen model (Qwen3.8 Flash Next GSQ RCO IQ3XXS) on an RTX 4080 with 16GB VRAM + 64GB DDR5 via strata and opencode (95 tps decode, 3000 tps prefill) to replicate the motion-graphics video trend popularized by Opus 5.5.
The recipe:
- Explicitly instruct the model to use three.js for the scene, Playwright for frame-by-frame rendering, and ffmpeg for MP4 encoding
- Sample prompt: "Make a 15-second, 1920×1080, 60fps video in Three.js that showcases the famous japan shinkansen train, like a showreel. Render it frame by frame to MP4 with Playwright and ffmpeg, go all out, 1s hook, loop."
- First render took 40 minutes; about 3 follow-up prompts fixed dark lighting, missing buildings, and excess text — roughly 2 hours total
- The stack is well-suited for iterative fixes, and video length is unlimited
Verdict: not quite Opus quality, but good enough for a train-themed short for his nephew; he plans to try Qwen 4 next.
More from coding & agent
- AWS case study: DSPy RLM mines years of Slack history into an AI IT help desk knowledge base — lateinteraction · 2026-11-09
- McKinsey: multi-step agents burn 5-30x more tokens than chat — ingliguori · 2026-11-09
- Apple ships official Xcode plugin for Claude Code, plus an AI playbook to find and build iOS apps — udmrzn · 2026-11-09
- Deaf developer builds Signal Dog, an AI agent plugin that hears door knocks and alarms — Scobleizer · 2026-11-09
- Google open-sources Elephant-Goldfish plugin: design with context, review with fresh eyes — Saboo_Shubham_ · 2026-11-09
- 2026 Linux Kernel Module Programming Guide goes open access with QEMU sandbox practice — burkov · 2026-11-09