Last week, the Qwen team at Alibaba unveiled its latest artificial intelligence (AI) model for image generation, known as Qwen VLo. Serving as the successor to the Qwen 2.5 vision language model, the new offering comes with a series of enhancements that elevate its capabilities. The Qwen VLo model facilitates both text-to-image and image-to-image generation, supporting user inputs in various languages including English and Chinese. In addition to generating images, the model possesses the ability to edit images both inline and on inputs provided by users.
Qwen VLo Accepts Prompts in Multiple Languages
In a recent announcement via a post on X (formerly Twitter), the Qwen team shared details about the new model, technically referred to as Qwen3-235B-A22B. Users can access the model through the company’s chat interface free of charge, with the option to use it without logging in.
Tests conducted by staff members of Gadgets 360 revealed that Qwen VLo is competitive with Google’s Imagen 2 in terms of image generation capabilities. While its instruction adherence and output quality rank slightly below Imagen-3 and OpenAI’s GPT-4o-enhanced features, it excels in generation speed and boasts a more favorable rate limit.
According to information on its GitHub page, the Qwen VLo model includes enhanced image comprehension, which allows for more precise inline edits without compromising the original structure of the input images. This improvement contributes to a higher overall quality in the generated outputs. Furthermore, the model demonstrates improved understanding of ambiguous and open-ended prompts, facilitating an output that aligns closely with user expectations.
Additionally, the Qwen VLo can engage in tasks related to image annotation, such as edge detection, segmentation, and prediction mapping. The company has also indicated that future iterations of the model will incorporate capabilities to accept multiple images as input, allowing for user-requested combinations.
Text rendering has seen enhancements, permitting the generation of accurate text in various fonts during testing of the model. The Qwen VLo is capable of processing images with dynamic aspect ratios, including extreme ratios like 4:1 and 1:3. Plans are underway to introduce functionalities that will enable the generation of images in disparate aspect ratios soon.
