Alibaba introduced a new AI model for working with images called Z-Image. It has a size of six billion parameters and is suitable for running on local devices. The model supports creating images with a resolution of up to 2K and allows for complex editing using text commands. You can test the new model here .
One of the key features of Z-Image is the “Prompt Enhancer” function, which helps the model better understand complex or vague user instructions. The model demonstrates photorealism, accurately conveys natural lighting, skin texture, depth of field, and color balance. It can simultaneously change facial expressions, environment, and lighting while maintaining the integrity of the image.
Z-Image has deep semantic and cultural understanding. The model is aware of landmarks, people, holidays, poetry, and other concepts, allowing it to create images with cultural context in mind. For image editing, the “Z-Image-Edit” function is available, which supports complex text commands.
According to the Elo Human Preference Assessment on the AI Arena platform, Z-Image shows high competitiveness among open image models.
