Tongyi Wanxiang 208

Tongyi Wanxiang offers text-to-image and image-to-image capabilities that can assist humans in image creation, significantly lowering the barrier to image design. In the future, it can also be applied to scenarios such as artistic design, gaming, and cultural & creative projects. Tongyi Wanxiang primarily includes three core functions: text-to-image generation, similar image generation, and image style transfer.

Tongyi Wanxiang is a large ai painting model from Alibaba Cloud's Tongyi family. It is built on Alibaba's in-house developed compositional generation model, Composer, with 5 billion parameters, and is trained on billions of text-image pairs. Through a diffusion-model-based "compositional generation" framework, Composer is able to decompose and recombine image design elements such as color palette, layout, and style, achieving highly controllable and extremely flexible image generation results.

Tongyi Wanxiang Features:

1. Text-to-Image: The text-to-image page has a clean interface on the left, with a text input box at the top where you can enter your prompt. Below the text box, 8 optional visual styles are provided, allowing you to choose from styles such as watercolor, oil painting, Chinese painting, flat illustration, anime, sketch, 3D cartoon, and more. After entering a text description and selecting a style, the ai will automatically generate the image. For convenience, the output image aspect ratio can be selected from three preset options: 1:1, 16:9, and 9:16.

2. Similar Image Generation: Users only need to provide a reference image to obtain AI paintings with similar content and style. Please note that the uploaded image must be under 10MB in size, and the supported formats include common ones such as JPG, JPEG, PNG, and BMP. Images generated by Tongyi Wanxiang can be directly used to create similar images — simply click the generate button, and 4 similar images will appear on the right side for download.

3. Style Transfer: The style transfer interface allows you to input two images: one as the original image and another as the target style image. The generated image will retain the content of the original image and the style of the style reference image. You only need to upload the original image whose style you wish to transform and a target style reference image, and the original image will quickly be processed into a creative image in the target style. As with similar image generation, the original image and style image must each be under 10MB, with the same supported formats.

Whether it's text-to-image or image-to-image, Tongyi Wanxiang has delivered a great surprise in terms of semantic relevance, image completeness, and level of detail.

Tongyi Wanxiang Advantages:

• Fast generation speed.

• Diverse text-to-image styles: Supports 8 styles, each with distinctly different characteristics and features. Whether it's anime style or realistic style, the performance is impressive.

• High similarity match between the generated image and the original: The product precisely matches the generated images with the original, preserving the characteristics and details of the source image. There's no need to worry about the generated image differing too much from the original.

• Style transfer preserves the original image information: Wanxiang retains the information of the original image, allowing the generated image to possess a new artistic style while still preserving the features of the source image.

The launch of Tongyi Wanxiang marks Alibaba Cloud's continued innovation in the field of artificial intelligence. At last year's World Artificial Intelligence Conference, the "Tongyi" large model family from Alibaba Cloud made its debut. Since then, models and products such as Tongyi Qianwen and Tongyi Tingwu have been rolled out one after another. With the unveiling of the new family member Tongyi Wanxiang, the three major directions of AI have now been fully connected. This means that Alibaba Cloud has the capability to process or generate text, speech, and image modalities within the large model domain.