Process — annie lee

Hi, I’m Annie -
I’m tech-driven designer specialized in data visualization and branding ->

IDEATION

We noticed a missing element in typical sake labels—how much can we infer about flavors, aromas, and ideal drinking temperature just from the design?

Are there patterns or emerging trends in sake branding?

LABEL GENERATION MODEL

Fine-tune Stable Diffusion XL (SDXL) with DreamBooth using LoRA (Low-Rank Adaptation) for local usage.

The model is trained with 3O images and adjusting prompts manually with temperature, flavor, and aroma data.

LABEL GENERATION MODEL OUTPUT

PAIRING RECOMMENDATION MODEL

Using BART, a transformer model from Hugging Face, this system predicts food pairings based on sake descriptions by learning next-sentence relationships from a curated dataset. The dataset includes flavor-food pairings we collected from the Sakenomy website and data generated from ChatGPT.

SAKE.AI IN ACTION

Input →

Floral Sake with Sweet Characteristics, Best served at room temperature

BART Output:

Light tempura, fresh sashimi, vegetable dishes

SDXL Output:

Next Steps

→ Web scrape Sakenomy for more data to generate more diverse food pairings.

→ Build a label generation tool that visualizes the flavors and drinking recommendations like temperature and food pairings.

← Back • Next →