A Preliminary Study on the Generation and Training Method of “Blank Space” Imagery in Architectural Design under AIGI Technology: Taking Suzhou Museum as an Example.
Only this pavilion has nothing, and one can sit and observe the panoramic view of the sky· Su Shi of the Northern Song Dynasty.
In recent years, the optimization role of Generative Artificial Intelligence (AIGC) technology in the early stages of architectural design has been increasingly valued, but there has been less research on the discussion of architectural generative design for a specific image. As a special architectural art technique, “blank space” contains profound Eastern aesthetic ideas in its architectural imagery, making it difficult to be accepted by generative algorithms through prompt keyword descriptions. Taking Su Bo as a typical case of blank space architecture, taking the process of learning Su Bo’s “blank space” imagery and training its Lora style model using the most widely used model, Stable Diffusion, as an example, this study explores the generation and training methods of “blank space” imagery in architectural design, which may provide some reference for using AIGC technology tools to generate and optimize special imagery in early design..
Artificial Intelligence Generated Content (AIGC) is an important milestone in the transition from the 1.0 era of artificial intelligence to the 2.0 era. Unlike the past analytical AI, the booming generative AI in recent years not only extracts information and predicts trends through learning data, but also generates new content that is different from learning samples through the continuous improvement of its parameters and training volume, bringing great convenience and imagination space to life and production..
1.2 Application Status of AIGI Technology in Image Generation in Architectural Design.
As a powerful branch of artificial intelligence, Generative Artificial Intelligence (AIGI) technology is also rapidly penetrating the field of architectural design, affecting the methods and processes of architect design work. In the early stages of architectural design, Generative Artificial Intelligence (AIGC) technology has a good optimization effect [4], especially for complex architectural forms with aesthetic shapes. By combining images and generative algorithms, complex forms with strong sense of form and even rich structure can be generated..
Generative artificial intelligence can learn to establish information processing methods that are consistent with the architect’s thinking through analogy, construct cognitive strategies, and ensure the implementation of model universality for some architectural image features that can be simply described in common languages, such as color, scale, material, and scenery. Compared to parameterized design and algorithm design, which first construct clear rules in the design process, this path of “fitting rules backwards from the results” will become more universal in dealing with various common generation tasks as the training data expands in type and quantity [6]. However, although the universal model can widely learn architectural system knowledge, generate images that match prompt words [7], and has strong generalization ability, it can achieve weight reuse and transfer when introduced into the field of architectural design. However, identifying special buildings.
