Qwen Image 2.1 Turbo and Qwen Image 2.1: Generate Now

It is a 7B parameter text-to-image generator released on 2026-09-20 by Alibaba's Qwen team. According to Alibaba, it achieved a score of 60.28 on their leaderboard, placing it first among open-weight models. It allows you to create and edit images simply by using everyday language. Its 8-step sibling, Qwen Image 2.1 Turbo, followed on 2026-10-09.

The weights are for research use only, and commercial use of the weights falls outside the license, while images you generate on RiceBowl can be used commercially. The BF16 version needs about 14GB of GPU memory, and the INT8 version needs about 7GB or more. To run it locally, you need to install ComfyUI, download the weights, and set up a workflow.

You can start using it without paying upfront. New accounts get 8 free credits on sign-up, no card required. Each generation you create uses credits from your account balance.

Yes, any images you generate on RiceBowl can be used commercially for your projects or business materials. However, if you want to use the actual model weights for commercial purposes, you need a separate agreement with Alibaba.

The model was developed and released on 2026-09-20 by the Qwen team at Alibaba. They placed the weights on Hugging Face strictly for research purposes.

Using the model here is straightforward because everything runs directly in your web browser. You just type a descriptive prompt, optionally upload your reference images, and click the button to process your request. The final picture appears in your feed.

Yes, the Qwen Image family allows for natural-language editing of your visuals. You can upload an existing picture and type out exactly what you want to change in the scene. You describe the change and the model edits the picture.

On RiceBowl, the uploader takes up to 8 reference images to guide your output. Alibaba's own examples show the model merging up to 10 reference images, such as grouping 6 individual portraits into one photo or combining 5 reference images into one outfit.

Yes, putting readable text inside images is a widely known strength of the Qwen Image family. You can write your prompts in both Chinese and English to get specific words integrated into the output. Simply wrap your desired words in quotes within your prompt.

You can create many types of subjects, such as posters with text, edited photos, and pictures built from several references. By writing clear prompts and uploading relevant reference photos, you guide the visual style.

Qwen Image 2.1 Turbo is the accelerated version of Qwen Image 2.1 from Alibaba's Qwen team, released on 2026-10-09. It keeps the same 7B architecture but generates in 8 denoising steps instead of the 40-step default, so one picture takes far fewer sampling passes.

Both share the same 7B model family and both handle text-to-image and editing with up to 10 reference images. Qwen Image 2.1 Turbo adds a saved 8-step sampling schedule and runs with CFG 1, which makes each generation faster. The base Qwen Image 2.1 uses the longer 40-step default.

Yes. The Qwen Image 2.1 Turbo weights load in Diffusers and drop into the official Qwen Image 2.1 ComfyUI workflow. You still need to download the weights yourself and have a GPU with enough memory, so many people prefer to generate in a browser instead.

The Qwen Image 2.1 Turbo weights are free to download for research under the Qwen Research License, but commercial use of the weights needs a separate agreement with Alibaba. On RiceBowl, new accounts get 8 free credits on sign-up, and images you generate here can be used commercially.

RiceBowl does not host a separate Qwen Image 2.1 Turbo checkpoint. This page covers the Qwen Image 2.1 family, and when you press generate you create the picture with an available image model in your browser, with no install or GPU needed.