Artificial intelligence turns simple text into realistic building designs

Researchers from the Japan Advanced Institute of Science and Technology (JAIST) have developed a retrieval-augmented generation system for creating accurate architectural images from text descriptions.
The new system uses text prompts to generate structural sketches of buildings and refines them with detailed architectural elements from real building components, unlike previous models that often failed to capture accurate spatial details.
The step-by-step process mirrors real architectural practice: initial structural sketch, refining with detailed elements, and final rendering that matches the designer’s intent.
To test the framework, the team used specialized datasets consisting of images related to campus buildings, including arrangements of floors, windows, and entrances, achieving notable accuracy and realism in generated images.
The system received positive feedback from architectural students, scoring above 4 on quality metrics such as image realism, consistency with prompts, and architectural detail accuracy.
The development promises to make early-stage architectural design workflows more efficient by allowing rapid revisions in response to client feedback and by helping visualize multiple design options.
The work was published in the journal Frontiers of Architectural Research and was supported by the JST BOOST Program and the National Natural Science Foundation of China.

You may have missed