In the past few years, image generation AI technology has evolved dramatically, and it is also being used in design and conceptual visual production sites.
By using models such as Stable Diffusion and Midjourney, beautiful visuals can be generated in seconds to minutes, and prototyping speed has been greatly improved.
But like “Architectural Perth”Images requiring structure, scale, and spatial consistencyHowever, there are still high hurdles to using image generation AI. In particular, from the viewpoint of “accuracy” and “reality” required at construction design sites, there are still several fundamental issues with the output of general image generation AI.
Common image generation AI uses natural languagePrompt inputThe image is output by While this “prompt” is extremely flexible and has a high degree of freedom, in areas where “specific dimensions,” “structure,” and “arrangement relationships” are emphasized, such as architecture,Ambiguity is a fatal weaknessIt will be.
For example, even if “LDK space in a wooden house with large south-facing windows” is entered into the prompt, problems such as “furniture arrangement that does not match the actual sense of dimensions,” “the number of walls does not match,” and “the way light enters is unnatural” frequently occur in the output image.
Also, there are many cases where the output content changes drastically even if the image is generated again under the same conditions,Ensuring reproducibility and consistency is difficultThere is also an issue called.
In order to avoid these problems, a very detailed prompt design is required. In other words, it is necessary to understand what kind of words to use to create what kind of visuals, and the characteristics of each AI model, and to make optimal descriptions for them.
However, this “prompt engineering” is not only about architectural knowledge,Knowledge and trial and error experience unique to AI toolsSince it is a prerequisite, it is associated with high learning costs for construction professionals.
As a result, generating architectural passages has become “a task that only some people who are familiar with AI can do,” and this is insufficient to meet the needs of designers to quickly confirm by themselves.
Based on these issues,ArchiexThen, we have adopted a “promptless (= no prompt input required)” approach. Instead of instructing with a prompt from scratch, it is a mechanism that can automatically generate an architectural perspective in line with the intention simply by uploading a reference image such as a sketch or elevation and indicating some details according to the options.
Since there is no need to make detailed adjustments at the prompt,Instantly generate accurate and consistent perspectives even for people who aren't familiar with AIIt can be done, and it will be possible to smoothly proceed with daily design studies and presentation material preparation.
In order to utilize the power of image generation AI in design practice, it is important to aim for “designs that anyone can use” rather than simply introducing tools. ARCHIEX will contribute to the future of architecture as a product that is the first step.
