~/LOCALLLAMA/motion-graphics-video-created-locally-using-qwen-and-three-js

Motion Graphics Video Created Locally Using Qwen and Three.js

A user generated a 60fps motion graphics video of a Shinkansen train using a locally run Qwen model combined with Three.js, Playwright, and FFmpeg. The entire rendering and revision process took about two hours on an RTX 4080 setup. This demonstrates a practical, local generative workflow for creating frame-by-frame motion graphics without duration limits, bypassing proprietary cloud video services. It highlights the growing capability of open-weight models when chained with standard web and media tools. The workflow used a Qwen model running at roughly 95 tokens per second decode speed on a 16GB VRAM GPU, instructing it to output Three.js code rendered via Playwright and compiled using FFmpeg. Iterative prompts were used to fix visual glitches such as darkness and missing background elements.

## BACKGROUND

Three.js is a popular JavaScript library used to create and display animated 3D computer graphics directly in a web browser using WebGL. Playwright is a web automation framework developed by Microsoft that can drive browsers for tasks like automated script execution and capturing web content, while FFmpeg is an industry-standard multimedia framework for recording, converting, and streaming digital audio and video.

## REFERENCES

## KEYWORDS

#LocalLLaMA#Generative AI#Three.js#Video Generation#Workflow

$ subscribe --daily