Skip to content
local-ai

Qwen-Image

Alibaba · Image generation · 20B

Commercial use permitted Open weights Apple Silicon

Alibaba's 20-billion-parameter image model, notable for strong text rendering and a permissive Apache 2.0 licence. It is designed to fit under 16GB with quantisation, and generates a 1024px image in a handful of steps.

Strengths

  • Strong at rendering readable text within images
  • Apache 2.0, so fully usable commercially
  • Fast, few-step generation, and designed to fit consumer cards

Weaknesses

  • A 20B model is heavy at full precision
  • A smaller fine-tune and LoRA ecosystem than SDXL
  • Newer, so local tooling and community resources are still maturing

Hardware requirements

QuantisationApprox. VRAMNotes
GGUF Q4~14GBQuantised, designed to fit under 16GB with offloading
FP8~20GBA good balance on a 24GB card

Optimised builds available for Apple Silicon.

Licence

Apache 2.0 read the licence

Availability

Recommended for

  • Commercial image work needing a permissive licence
  • Images that depend on accurate text
  • Users with a 16GB or 24GB card who want an alternative to FLUX

Catalogue entry last verified 30 July 2026. Specifications change; verify anything you are about to spend money on.