Open Source

Qwen3.6-27B-Q6_K - images

AI creates capybara in kimono, flamingo knitting — all as scalable vector graphics.

Deep Dive

A Reddit user posted performance stats for generating SVG images from natural language prompts: a pelican riding a bicycle, a capybara in a kimono drinking matcha, a flamingo knitting a sweater, a sushi roll in sunglasses driving a go-kart, a Victorian robot reading a newspaper in a cafe, and a time-lapse flower. Generation times ranged from 3 minutes 10 seconds to 8 minutes 24 seconds, with speeds of 27.55, 27.05, 27.55, 27.27, 27.19, and 27.13 tokens per second.

Key Points
  • Generates SVG images from text prompts like 'capybara wearing a kimono' and 'Victorian robot reading a newspaper'
  • Runs at ~27 tokens/second with generation times from 3 to 8.5 minutes per image
  • Uses Q6_K quantization (27B parameters) for efficient consumer-grade hardware deployment

Why It Matters

Enables direct vector illustration from text, offering scalable, editable outputs for designers and artists.