Qwen Open-Sources Qwen-Image 2.1

If you run image models on your own machine, there is a new one for the list. Qwen has open-sourced Qwen-Image 2.1, a 7B model that handles both generation and editing in one place. It is on Hugging Face now, and Diffusers already supports it.

That means no waiting on a community wrapper and no API bill. You pull the weights, point your setup at them, and the editing model is the same model that draws the image.

What Happened

• Qwen-Image 2.1 is open-sourced and published on Hugging Face: https://huggingface.co/Qwen/Qwen-Image-2.1
• It is a 7B parameter image model.
• Generation and editing live in the same model, not two separate ones.
• It produces native transparent PNGs.
• It accepts up to 10 reference images.
• It runs locally, and Diffusers already supports it.

Why It Matters

A transparent background usually means an extra step: generate the image, then cut the background out. Here the alpha channel comes from the model itself, so anything headed for a design tool skips the cleanup pass.

The 10 reference images are the other part worth noticing. Feeding more than one example gives the model something to match, instead of you trying to describe the look in text alone.

If you already have a local image setup, the weights are sitting there waiting. Start at https://huggingface.co/Qwen/Qwen-Image-2.1 and see where it fits in your own workflow.