Open weights

Qwen-Image

Alibaba (Qwen team) · Aug 4, 2025 · Apache-2.0 · site · source

20B MMDiT image foundation model from the Qwen team — Apache-2.0, exceptional at Chinese text rendering.

What it is

Qwen-Image is a 20B-parameter MMDiT image foundation model released by Alibaba's Qwen team on Aug 4, 2025 under Apache-2.0. Beyond text-to-image, the lineage includes a deep editing track (Qwen-Image-Edit, with 2509 / 2511 / Edit-2511 multi-person variants) and the refreshed Qwen-Image-2512 base (Dec 31, 2025). It's particularly strong at complex text rendering — especially Chinese — and supports native 2K output across 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, and 2:3 aspect ratios. Technical report: arxiv 2508.02324.

Method
Text → Image
Max res
2048×2048
Speed
Standard (5–15s)

Generated with Qwen-Image

No public samples for Qwen-Image yet.

Honest take

Strengths & limitations

Strengths
  • +20B parameters — largest permissively-licensed open DiT in this generation
  • +Best-in-class Chinese text rendering; strong English too
  • +Mature editing variants (2509, 2511, Edit-2511) with multi-person support
  • +Apache-2.0 — fully commercial-clean
Limitations
  • 20B is heavy — self-hosting needs serious VRAM vs Z-Image's 6B

Need stock imagery on a brief?

OKSLOP handles routing, provider selection, and delivery — you describe what you need and we hand back the shots.