New Image 2.0 model on Qianwen APP.

On March 13th, Qwen-Image2.0 image generation and editing model was launched on the Qianwen APP. According to reports, Qwen-Image 2.0 integrates image generation and editing, supports text-to-image generation, basic editing, and multi-scene creation, and has been fully upgraded in complex text rendering, realistic texture representation, and instruction understanding. Qwen-Image 2.0 was officially released on February 10th. This model supports 1K token text input and 2K high resolution rendering of complex instructions, and has the ability to render Chinese characters. In addition, a lighter model architecture has been adopted, with a model size much smaller than Qwen-Image 1.0's 20B, resulting in faster inference speeds. According to the Qianwen team, Qwen-Image 2.0 performed well in model blind tests conducted in the AI Arena, showing superior performance in both text-to-image and image-to-image benchmarks with the same model.