Qwen-Image-2.1: Editing, Transparency, Local Setup and License

What Qwen-Image-2.1 adds

Qwen announced Qwen-Image-2.1 on September 20, 2026. The official post describes a unified image-generation and editing model with a 7B visual-generation component, native transparent-image creation and editing, and support for local edits. These are upstream capabilities. PromptToVisual does not currently host Qwen-Image-2.1, and our listed Qwen Image 3 Pro option is a different model.

Generation and editing live in one model

The official release combines text-to-image and reference-image editing rather than publishing separate generation and edit checkpoints for these tasks. Qwen says the model can use up to 10 reference images, preserve people and products, extract subjects, and work with transparent images. A maximum reference count is not a promise that every ten-image prompt will preserve every identity or detail. Inspect each result against the references.

Transparent output is a real upstream feature

The release specifically calls out generating regular or transparent images and editing transparent layers. That is useful for cutouts, overlays and design assets, but an RGBA result still needs inspection around edges, semi-transparent regions and small details. Do not assume that transparency automatically means a production-ready product cutout or perfectly isolated hair, glass or reflections.

How to run it locally

The official Hugging Face card documents Diffusers loading with Qwen/Qwen-Image-2.1 and a QwenImage21 pipeline path. Start from the current official README rather than copying an older Qwen-Image command. Record the exact model revision, precision, device setup and output dimensions if you are comparing local runs. We have not installed or benchmarked this model locally, so this page is source-based rather than a hardware performance test.

What about ComfyUI?

The official sources reviewed for this guide document the model and Diffusers workflow; they do not establish an official Qwen-Image-2.1 ComfyUI graph. Community nodes or workflows may exist, but a third-party graph should be evaluated separately for checkpoint compatibility, required custom nodes and license terms. Do not describe a community workflow as an official same-day Qwen integration without evidence.

The license is the important commercial boundary

The published Qwen Research License defines the granted use as non-commercial research or evaluation. It states that commercial use requires a separate commercial license from Qwen. That means “open weights” is not the same as unrestricted commercial use. Check the current license file yourself before putting the weights behind a paid product, API or client workflow. This guide is not a legal opinion.

When should a PromptToVisual user care?

Use this guide when you are evaluating Qwen-Image-2.1 itself, especially for local research, transparency or multi-reference editing. If you need a hosted image workflow now, use one of the models actually listed in the PromptToVisual studio and follow that model page’s displayed cost and input limits. Selecting Qwen Image 3 Pro here does not secretly run Qwen-Image-2.1.

Sources and next steps

Source check: September 26, 2026. Qwen announced the model September 20. The official weights are under the Qwen Research License, which limits the granted use to non-commercial research/evaluation unless a separate commercial license is obtained. PromptToVisual does not host Qwen-Image-2.1.

Qwen: Qwen-Image-2.1 announcement →

Hugging Face: Qwen-Image-2.1 model card →

Qwen Research License →

PromptToVisual Qwen Image 3 Pro option → · Compare exported results → · All guides →

Continue with the right tool

AI image combiner · Stitch images · Photo collage · Nano Banana 2 guide · Related composition guide