When was Qwen Image 2.1 released?
The Qwen Research License Agreement shipped with the model carries a release date of 20 September 2026, and the release was reported the same day. The Hugging Face repository itself was created on 14 September 2026, ahead of the public announcement.
| Date | Event |
|---|---|
| August 2025 | Original Qwen-Image released under Apache 2.0 |
| 14 September 2026 | Qwen/Qwen-Image-2.1 repository created on Hugging Face |
| 15 September 2026 | Comfy-Org/Qwen-Image-2.1 repository created for ComfyUI files |
| 20 September 2026 | Public release; license dated; prompt-enhancement models (PE-T2I, PE-I2I) published |
| From 20 September 2026 | Community GGUF, FP8 and MLX conversions appear on Hugging Face |
What is new in Qwen Image 2.1
The model card names four improvements for this release:
- Compact and efficient: 7B parameters in the visual generation component, built from 32 single-stream DiT layers with mixed-granularity attention and prefix KV cache reuse.
- Native transparency in a unified model: generate regular or transparent RGBA images from text, edit transparent layers, and extract subjects from photographs with the same weights.
- Versatile editing: up to 10 reference images, local edits specified with circles, painted annotations or separate masks, and identity preservation for people and products.
- Realistic textures and refined aesthetics: improved typography, portrait lighting and fine detail.
The model generates natively at 2K. The card lists seven sizes from 2048 x 2048 (1:1) to 2752 x 1536 (16:9) and 1536 x 2752 (9:16).
Is Qwen Image 2.1 open source?
The weights are open: anyone can download them from Hugging Face or ModelScope without a gate, and the model card uses the word open-source. The license, however, is not a permissive open-source license.
Qwen-Image-2.1 is covered by the Qwen Research License Agreement. It grants the right to use, reproduce, modify and distribute the model for non-commercial purposes only, and defines non-commercial as research or evaluation purposes. Commercial use requires a separate commercial license from the Qwen team.
| Model | License on Hugging Face | Commercial use |
|---|---|---|
| Qwen-Image (August 2025) | Apache 2.0 | Allowed under Apache terms |
| Qwen-Image-2.1 (September 2026) | Qwen Research License | Requires a separate license |
- Redistributed copies must include the agreement and the specified attribution notice.
- Models trained or improved with the materials or their outputs must display "Built with Qwen" or "Improved using Qwen" in their documentation.
- Derivatives may not use "Qwen" as their primary name, though descriptive use is permitted.
This is a plain-language summary, not legal advice. The LICENSE file in the official repository is the binding text.
Where you can use Qwen Image 2.1 today
- Official demo: the Qwen/Qwen-Image-2.1 Space on Hugging Face.
- Locally: the QwenImage21Pipeline in Diffusers, or the native ComfyUI templates for text-to-image, image edit and background removal.
- Lower-memory setups: INT8 files from Comfy-Org and community GGUF quantizations.
- In the browser here: this studio runs Qwen Image 2.1 generation and editing through a hosted API, for 1 credit per image.
qwenimage21.pro is an independently operated workspace. It is not affiliated with or endorsed by Alibaba or the Qwen team, and it does not host or redistribute model weights.