PUZZLE WOOD Scent Data Visualization
Summary
Rather than expressing the inspiration or emotions evoked by one’s subjective perception of a scent,
this project seeks to visualize the inherent characteristics and objective
data of fragrance through carefully constructed visual imagery.
▲ PUZZLE WOOD Vegan Diffuser
To visualize the objective attribute data of
the Puzzlewood fragrance using generative AI
tools, the process was carried out in four stages
Step 1: Collect and analyze fragrance data
Step 2: Develop data algorithms
Step 3: Visualize the data
Step 4: Generate images through generative AI using the graph-based visuals
the Puzzlewood fragrance using generative AI
tools, the process was carried out in four stages
Step 1: Collect and analyze fragrance data
Step 2: Develop data algorithms
Step 3: Visualize the data
Step 4: Generate images through generative AI using the graph-based visuals
Step 1: Collection and Analysis of Puzzle Wood Fragrance Data
We reviewed a range of materials related to the Puzzle Wood fragrance and decided to use the MSDS (Material Safety Data Sheet) as it serves as a comprehensive source of fragrance information.
The MSDS provides detailed information necessary for the safe handling and management of chemical substances. (As this data contains proprietary product information, high-resolution images will not be shared.)
Step 2: Data Algorithm Development
From the various information listed in the MSDS, we selected the chemical names of the components, CAS numbers, and fragrance concentration data.
Information presented in different formats and types—such as text, numerical sequences, and ranges—was converted into a unified numerical format.
Chemical names expressed in English text were converted into numerical values using ASCII codes, while CAS numbers, which consist of three groups of digits,
were standardized by multiplying the numbers within each group to align their digit ranges.
Specific rules were then applied to each dataset, creating a customized algorithm that enabled the data to be expressed in a consistent and unified structure.
▲ [Numerical conversion of three data types based on Puzzlewood fragrance MSDS information]
Step 3: Data Visualization
◀ [Inputting numerically converted data into the generative AI DALL·E to create images]
▲ [Inputting numerically converted data into the generative AI DALL·E to create images]
Next, we visualized the data that had been organized numerically. Initially, we entered the numerical data into the generative AI DALL·E to generate images.
However, as the work progressed, it became difficult to verify whether the resulting visuals were truly derived from the data or simply based on the descriptive prompt.
Consequently, we changed our approach.
Instead, we adopted a more conventional method of data visualization—using graphs.
Since the dataset consisted of random numeric sequences rather than trends or temporal changes, we determined that a scatter plot would be the most suitable format.
After researching programs that allowed for variations in point color and size, we selected MATLAB, an engineering software.
Chemical names were mapped to the X and Y coordinates, CAS numbers were converted into RGB values to define the color of each circle, and concentration data was applied to the circle size.
During this process, we further refined the dataset. To use CAS numbers as RGB values, we divided the integer data by 256 and used the remainder to convert them into integers ranging from 0 to 255.
To adapt this data for MATLAB, we then linearly transformed the values into floating-point numbers between 0 and 1. Through this series of complex steps, we generated our first visualized output.
▲ [Using all elements and delimiters of the converted numerical data to assign meaning to each point, generating a scatter plot with MATLAB]
Step 4: Image Generation with Graph-Based Input
In the final stage, we imported the scatter-plot image into the generative AI Midjourney to create new visuals.
We prompted it to preserve the colors and positions of the original points while filling the remaining space to depict Puzzlewood and its fragrance.
However, the results initially remained simplistic—limited to dots, trees, and perfume-bottle-like forms.
To move beyond such literal imagery, we refined the prompts to retain the mood, colors, and structure of the original scatter plot while evoking the sensation of fragrance diffusion.
After multiple iterations, we achieved a final image that successfully expressed this concept. Subsequently, we imported
the Midjourney-generated image into Runway to add animation, enhancing the sense of the fragrance’s movement and diffusion in a more dynamic, lifelike manner.
▲ [Input the scatter-plot image into generative AI Midjourney to create visuals, then use Runway to produce motion video]
Closing Remarks
At first, I simply wanted to visualize objective quantitative data as images, assuming that generative AI could create anything.
However, throughout the process, I came to realize that producing visually compelling results from raw data is not something achievable
with a basic image-generation tool alone—it requires deep-learning and machine-learning techniques as well as specialized programming.
As someone without a background in statistics or engineering, every part of this journey—from developing algorithms and writing code to using professional software—was a new experience.
It took significant time and effort, but ultimately, the process of visualizing objective fragrance data in my own way proved deeply meaningful.
- Amorepacific Creatives
- Planning
- 여지예
