Tongyi Wanxiang

Official homepage entrance, invitation code for Tongyi Wanxiang, app address

What is Tongyi Wanxiang?

Following Tongyi Qianwen, Alibaba Cloud's Tongyi large model family welcomes a new member. On July 7th at the 2023 World Artificial Intelligence Conference, Alibaba Cloud announced the launch of theinvitation-only beta testing for the AI painting creation large model Tongyi Wanxiang. This model assists humans in image creation and can be applied to art design, e-commerce, gaming, and cultural and creative industries in the future. Alibaba Cloud's CTO Zhou Jingren said this is a key step for Alibaba Cloud's large models to fully master multimodal capabilities, which will gradually be opened to industry clients.

Tongyi Wanxiang official website:https://wanxiang.aliyun.com/

How is Tongyi Wanxiang?

Tongyi Wanxiang excels in semantic understanding and image generation. The first batch of three capabilities include: in the basic text-to-image function, it can generate images in styles such as watercolor, flat illustration, anime, oil painting, Chinese painting, 3D cartoon, and sketch based on text content; in the similar image generation function, after uploading any image, users can create AI paintings with similar content and style; the model also leads in image style transfer, where users can automatically process the original image into the specified style by uploading the original image and style image.

Tongyi Wanxiang is based on Alibaba's developed composite generation model Composer, which can provide highly controllable and highly free image generation effects by decomposing and combining image design elements such as color matching, layout, and style. It can support multiple image generation tasks with just one model, and this research has been published at the international AI top conference ICML 2023.

Alibaba Cloud demonstrated the image generation effect of the model on-site. When inputting "Immersed in a captivating world of full-page grayscale painting, with a tiger in the tranquil forest of mandala, the scene uses lines and strokes," the system quickly generates a tiger image with rich line details. If the style transfer function is continued to be used, the image style can be freely switched.

At last year's World Artificial Intelligence Conference, Alibaba Cloud's "Tongyi" large model series made its debut, followed by the launch of models and products such as Tongyi Qianwen and Tongyi Tingwu. The release of Tongyi Wanxiang means that Alibaba Cloud has the ability to process or generate text, voice, and images in the large model field. Currently, Alibaba Cloud's large models are gradually moving towards various industries, with over 300,000 enterprises applying to test Tongyi Qianwen, and Tongyi Tingwu has accumulated 360,000 users.

"Multimodal is an inevitable path for the evolution of large models. We hope to use the capabilities of different modalities to serve thousands of industries and help enterprises fully embrace the intelligent era," Zhou Jingren said.

Alibaba Cloud introduced at the conference that Tongyi Wanxiang has the capabilities of text-to-image and image-to-image, which can assist humans in image creation, significantly reduce the threshold for image design, and can be applied to art design, gaming, and cultural and creative industries. The model has already startedinvitation-only beta testing, and the specific functions launched in the first batch include the following three:

• Text-to-Image GenerationInput a description of the scene in text and select a creative style (watercolor, oil painting, Chinese painting, anime, etc.), and AI can automatically generate the creation image.
• Similar Image GenerationProvide a reference image to obtain AI paintings with similar content and style.
• StyleTransferUpload the original image and style image, and the original image can be automatically processed into the specified style.