story A busan x Audio Book Exhibition
Summary
Assuming the planning of an audiobook exhibition for Story A located in Haeundae, Busan, this project aimed to explore the use of various AI tools throughout the entire exhibition planning process
— from generating concept ideas to creating spatial reference images and visual sources for spatial design.
▲ PROMPT :an audiobook exhibition, small space, modern and simple mood, dimly lights, asian woman, closing eyes, black color beanbags, dark room.
Concept
How many hours in a day do we truly spend away from our computers, mobile phones, and tablets, aside from when we sleep?
When was the last time you became so absorbed in a book that you escaped the endless stream of visual stimuli?
What if, instead of watching, we closed our eyes, listened, and let our imagination take over?
We invite you to experience the story of Busan, where the sound of waves feels almost within reach.
AI Tool
Chat GPT
Midjourney
Stable Diffusion
Firefly
vectorizer.ai
Midjourney
Stable Diffusion
Firefly
vectorizer.ai
1. Contents
Recently, I found myself staring at my phone to the point of feeling dizzy, unable to look away from the constant notifications and the endless stream of stimulating videos.
Even at night, I often stayed up watching them until I could no longer fight sleep, only to struggle to wake up the next morning.
I tried countless times to read a book instead, but putting down my phone and truly focusing on reading turned out to be far more difficult than I expected.
When brainstorming new exhibition themes related to culture and art for Story A, it felt natural to think, “What if there were a space where people could put down their phones and listen to books—like being in a movie theater?”
Leveraging the unique identity of Busan, I decided to plan an audiobook exhibition that tells the stories of the city. With this basic idea in mind, I turned to Chat GPT to further enrich the concept and make the space more dynamic—starting with a few simple questions in hopes of receiving unique and unexpected ideas.
Reading Chat GPT’s responses helped me refine and expand on my ideas, even gaining inspiration in areas I hadn’t previously considered.
However, when I repeated similar questions, the answers became somewhat redundant—only slightly rephrased versions of earlier responses.
Later, I came across a review suggesting that engaging Chat GPT with encouragement might lead to more creative outcomes, though I didn’t realize that at the time.
After shaping the concept, I asked for more concrete information—such as novels set in Busan, authors from Busan, and examples of audiobook exhibitions.
Unfortunately, much of the information provided turned out to be inaccurate, and verifying it took more time than expected, leaving me with little usable material.
In conclusion, Chat GPT proved useful for refining language, generating ideas, and developing conceptual directions, but it was not yet reliable for sourcing fact-based information.
2. Space Design
My motivation to explore AI began with a somewhat ambitious thought: “Since AI is trending, if I learn how to use it well, maybe I won’t need to struggle with SketchUp anymore—AI can handle everything for me.”
To create spatial references and conceptual visuals, I provided prompts describing my envisioned scenes to Midjourney and Stable Diffusion to generate images.
While their speed and rendering performance were truly impressive, the process of achieving the exact image I wanted was far from simple.
▲ PROMPT : an exhibition room, dimly lights, lying people on black beanbags, audio exhibition,
audio-visual installations, fujifilm eterna vivid 500t, reimagined by industrial light and magic
audio-visual installations, fujifilm eterna vivid 500t, reimagined by industrial light and magic
▲ PROMPT : an exhibition room, dimly lights, lying people on black beanbags, audio exhibition,
audio-visual installations, fujifilm eterna vivid 500t, reimagined by industrial light and magic
▲ PROMPT : an audiobook exhibition, small space, modern and simple mood, dimly lights, asianm
woman, closing eyes, black color beanbags, dark room
woman, closing eyes, black color beanbags, dark room
▲ PROMPT : an audiobook exhibition, small space, modern and simple mood, dimly lights, asianm
woman, closing eyes, black color beanbags, dark room
Midjourney left a strong impression on me with its ability to generate infinitely varied images even when using the same prompt repeatedly—each maintaining a similar mood but depicting completely different scenes.
It primarily produced dreamlike and fantastical imagery, which often felt otherworldly.
What fascinated me most was how these unfamiliar scenes continuously sparked new associations and visual ideas, ultimately inspiring fresh concepts for spatial design.
However, due to my still-developing prompting skills, it was difficult to create images that perfectly matched the scenes I had envisioned.
For instance, when I entered “black color,” the output often shifted to amber or dark green tones, and prompts like “Asian girl” frequently generated images of Latina-looking women, which was somewhat frustrating.
▲ PROMPT : an exhibition room, dimly lights, lying people on black beanbags, audio exhibition,
audio-visual installations, fujifilm eterna vivid 500t, reimagined by industrial light and magic
audio-visual installations, fujifilm eterna vivid 500t, reimagined by industrial light and magic
▲ PROMPT : an exhibition room, dimly lights, lying people on black beanbags, audio exhibition,
audio-visual installations, fujifilm eterna vivid 500t, reimagined by industrial light and magic
▲ PROMPT : an exhibition room, dimly lights, lying people on black beanbags, audio exhibition,
audio-visual installations, fujifilm eterna vivid 500t, reimagined by industrial light and magic
audio-visual installations, fujifilm eterna vivid 500t, reimagined by industrial light and magic
▲ PROMPT : an exhibition room, dimly lights, lying people on black beanbags, audio exhibition,
audio-visual installations, fujifilm eterna vivid 500t, reimagined by industrial light and magic
They say that a true craftsman never blames their tools—but in the case of Stable Diffusion, I’ll make an exception.
It absolutely requires a high-performance PC as a baseline.
Under that condition, it was capable of producing images at impressive speed, and compared to Midjourney, the results had a more realistic atmosphere.
Unlike Midjourney, which instantly generates images based purely on text prompts, Stable Diffusion demands a certain level of technical proficiency.
However, with enough practice, it becomes a tool that allows for more nuanced control and fine-tuning during the creative process.
I experimented by using a basic 3D spatial layout of Story A’s Busan site, exported from SketchUp, as a reference composition and inputted prompts to generate space renderings.
Still, achieving the exact visual I had in mind remained difficult. (See bottom-right image.)
For example, when I simply wanted to create a black wall and entered “black wall,” the result was sometimes an unsettling image—with a black figure embedded in the wall.
After experimenting with both tools, I found that creating precise and site-specific spatial designs using generative AI is still challenging. However, if you are at the conceptual stage, with only an abstract idea or mood in mind, tools like Midjourney or Stable Diffusion can be highly effective. They help you visualize vague ideas, and the unexpected outcomes often lead to new, unplanned creative directions.
After experimenting with both tools, I found that creating precise and site-specific spatial designs using generative AI is still challenging. However, if you are at the conceptual stage, with only an abstract idea or mood in mind, tools like Midjourney or Stable Diffusion can be highly effective. They help you visualize vague ideas, and the unexpected outcomes often lead to new, unplanned creative directions.
3. VMD Design
This time, assuming that the audiobook content—which currently exists only as audio files—would be transformed into an offline exhibition featuring physical books,
I experimented with generating book covers for short stories. The stories were planned around themes related to Busan, so my goal was to create a variety of images that evoked the sea,
beaches, and the cityscape. As before, I provided the same set of keywords to both Midjourney and Stable Diffusion to test their capabilities.
▲ Midjourney
▲ Stable Diffusion
Midjourney primarily produced dreamlike, fantasy-inspired images, with strong realism in its landscapes.
By adding the keyword “illustration” and switching between different versions, I was able to generate a wide range of design variations.
On the other hand, Stable Diffusion tends to generate images with similar styles depending on the checkpoint or LoRA used, so finding the right combination of these became the key to achieving stylistic diversity.
From the generated images, I selected those most suitable for book covers and adjusted their composition and proportions to fit the required layout. This process was made quick and efficient using Firefly.
Initial image creation was done through Midjourney or Stable Diffusion, and then Firefly was used to expand the canvas, fix awkward details, remove unwanted elements, and fine-tune the visuals.
This parallel workflow significantly reduced production time. However, achieving exactly the style I wanted was far from easy—patience was essential, as it often took numerous iterations before a satisfying result emerged.
Additionally, I wanted to convert the images into vector files that could be edited in detail and used for actual production without size limitations. The “Image Trace” tool in Company A’s “I” program didn’t work well, so I searched for alternatives and discovered a very useful website: vectorizer.ai. This tool allowed me to quickly and easily vectorize image files—even for free—and export them as .eps files. It’s particularly useful for turning realistic images into illustrations or extracting specific visual elements for reuse, so I’ve been recommending it to colleagues.
Additionally, I wanted to convert the images into vector files that could be edited in detail and used for actual production without size limitations. The “Image Trace” tool in Company A’s “I” program didn’t work well, so I searched for alternatives and discovered a very useful website: vectorizer.ai. This tool allowed me to quickly and easily vectorize image files—even for free—and export them as .eps files. It’s particularly useful for turning realistic images into illustrations or extracting specific visual elements for reuse, so I’ve been recommending it to colleagues.
▲ Basic image generated using Midjourney
▲ Vectorized version created using the vectorizer.ai site
Finally, I added the novel titles and completed the final file, ready for production.
There’s something curiously nostalgic about it — a subtle 1980s book cover vibe.
Although it’s not entirely to my satisfaction, I’m sharing the outcome as a way to conclude my AI challenge project.
Through this process, I learned that instead of relying on a single AI tool, effectively combining multiple tools according to each
situation can significantly reduce the time spent starting from scratch and minimize labor-intensive work — ultimately leading to production-ready results.
I look forward to the day when a positive collaboration with an audiobook platform becomes a reality, and the power of AI is truly brought to life within the Story A space.
- Amorepacific Creatives
- Planning & Design
- 백지영
