Qwen releases Qwen-Image-2.1 for unified image generation and editing

Archive entry: this covers an event from September 20, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.

On September 20, 2026, Qwen released Qwen-Image-2.1, a model that unifies text-to-image generation and image editing, as described in the Qwen-Image-2.1 README on GitHub. The release came with weights on Hugging Face and ModelScope and with day-0 support in Diffusers, ComfyUI and several inference engines.

What the model offered

• The visual generation component had 7B parameters across 32 Single-Stream DiT layers.
• It supported native RGBA transparency and up to 10 reference images.
• Local edits could be made through circles, annotations and masks.
• Weights were released on the Qwen-Image-2.1 page on Hugging Face and on ModelScope under the Qwen Research License (qwen-research).

What it means for people who build with AI

Creators who use ComfyUI or Diffusers could run an image generator and editor with released weights from day one. Because Qwen-Image-2.1 unified generation and editing, a single model covered text-to-image work as well as edits driven by reference images, circles, annotations and masks. Day-0 support reached beyond those tools to several inference engines and to Intel hardware through OpenVINO.

Day-0 support

• Diffusers supported the model through QwenImage21Pipeline.
• ComfyUI, vLLM-Omni, SGLang and LightX2V supported the model from release.
• Intel OpenVINO supported it on Arc Pro B70 and Core Ultra Series 3.

Sources

• Qwen-Image-2.1 (GitHub README) (Alibaba (Qwen), September 20, 2026, official)
• Qwen/Qwen-Image-2.1 (Alibaba (Qwen), official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.