Qwen-Image 3.0 🧠 Model Open source
Alibaba's Qwen image models, known for dense multilingual text rendering, with open-weight versions
- GitHub stars
- 8.4k
- Stars this week
- –
- Forks
- 554
- Licence
- Apache-2.0
- Last push
- 2026-02-10
- Maintainer
- Alibaba (Qwen team)
Where you can use Qwen-Image 3.0
Apps in this directory that let you generate with Qwen-Image 3.0, from their own documentation.
Good for
About Qwen-Image 3.0
What it is
Qwen-Image is Alibaba's image generation and editing family. The newest hosted model is Qwen-Image-3.0 (14 September 2026) on Alibaba Cloud Model Studio, after Qwen-Image-3.0-Pro (July 2026) and Qwen-Image-2.0-Pro (June 2026). On the open side, Qwen-Image-2.1 (7B visual generator, September 2026) joined the 20B Qwen-Image, Qwen-Image-2512 and Qwen-Image-Edit-2511 on Hugging Face.
Key features
- Complex text-and-image layouts from prompts of up to 4.5K tokens, with stable text in 12 languages (Qwen-Image-3.0)
- Dense layouts such as newspapers and exam papers, down to about 10-pixel text (3.0-Pro)
- Native transparent (RGBA) images and up to 10 reference images for edits (2.1)
- Output up to 2048x2048 in ratios from 1:1 to 9:16
- Identity-preserving edits and multi-person consistency in the Edit models
Where you can use it
Qwen Chat (free) and the Alibaba Cloud Model Studio API. Krea offers Qwen 2512, Image 2.1 and Image 3; ComfyUI supports Qwen-Image natively and Qwen Image 3 through partner nodes; InvokeAI supports Qwen Image and Qwen Image Edit locally. The weights are also on Hugging Face and ModelScope.
Pricing and rights
The original Qwen-Image and 2512 weights are Apache 2.0, so commercial use is allowed. Qwen-Image-2.1 is under the Qwen Research License; read it before commercial use. Qwen-Image 3.0 is API-only, billed per image on Alibaba Cloud; we did not find a public per-image price to quote. There is no paid consumer plan.
Who it is for
Designers making text-heavy graphics such as posters, slides, menus and infographics, especially in Chinese or multiple languages, and ComfyUI users who want an open alternative to FLUX.
Verdict
Qwen-Image is a leading choice for text in images, and its older open weights are genuinely permissive. The best new versions are cloud-only, and licensing differs from one release to the next.
Pros
- Renders dense text layouts, down to about 10-pixel text on Qwen-Image-3.0-Pro
- Stable text rendering in 12 languages, including Chinese
- Open weights: Qwen-Image (20B) and 2512 under Apache 2.0
- Qwen-Image-2.1 adds transparent (RGBA) output and up to 10 reference images
Cons
- Qwen-Image 3.0 and 3.0-Pro are cloud-only on Alibaba Cloud
- Qwen-Image-2.1 weights use the Qwen Research License, not Apache 2.0
- Few mainstream creative apps offer it outside Krea and ComfyUI
- 20B open model needs a large GPU or quantisation
Similar AI models
All image models →GPT Image 2.5 🧠 ModelFreemium
OpenAI's image model behind ChatGPT Images, with 4K output and up to 16 reference images
Nano Banana 2 🧠 ModelFreemium
Google's Gemini image models (Nano Banana 2, Pro and 2 Lite) for generation and editing up to 4K
FLUX.2 🧠 ModelOpen source
Black Forest Labs' FLUX.2 family: hosted [max]/[pro]/[flex] plus open-weight [dev] and [klein]
Midjourney V8.2 🧠 ModelPaid
Midjourney's own image model, known for aesthetics and personalization, now in V8.2
Seedream 5.0 Pro 🧠 ModelFreemium
ByteDance's Seedream image models, used in CapCut, Dreamina and many creative apps
Ideogram 4.0 🧠 ModelOpen source
Ideogram's text-rendering image model with JSON prompting, now also as non-commercial open weights