Turning Mixed Drinks into Visual Moods & Micro-Stories

A system that uses a live camera to capture the process of mixing a drink and transforms the recipe—its ingredients, colors, and proportions—into a generative visual or micro-story that reflects the drink’s mood, with a final downloadable/ shareable picture or video as a personal “drink memory.”

Inspiration

I often mix drinks with friends or by myself, and I want a beautiful way to remember the recipes we create. I’m not good at photographing drinks, and even when I do, photos rarely capture the mood or personality of the moment. I began imagining drink-mixing as a small performance: the colors swirling, layers forming, and ingredients interacting. This project grows out of the desire to preserve those moments—not as literal photos, but as expressive visuals or short stories that reflect the emotional tone of the drink being created.

Project Summary

This project explores how a physical drink-mixing process can be translated into real-time visuals and narrative. Instead of typing ingredients or selecting from a menu, the user places a glass in front of a camera and starts pouring. As liquids enter the glass, the system uses computer vision to identify the type of drink being added and estimate its relative amount. These signals drive a generative visual or a micro-story that embodies the “mood” of the mixture.

The core concept is to treat mixing as performance. A live camera continuously captures the glass, segmenting the drink region and analyzing color changes, opacity, and fill level. A pre-trained drink-recognition model estimates which type of liquid is present (e.g., juice, soda, coffee, spirits, milk-like beverages) using color and texture cues. A second model or heuristic estimates volume by tracking liquid height over time.

From these readings, the system constructs a dynamic recipe vector—a multi-dimensional representation of properties like sweetness, bitterness, acidity, creaminess, alcohol strength, and temperature association. This vector maps directly into a mood profile, controlling generative visual parameters such as color palette, contrast, particle speed, diffusion, and rhythm. The same parameters also guide the narrative tone of an accompanying micro-story.

For example:

• A bright, citrus-heavy drink might produce quick, airy particles in warm colors, paired with a story that evokes lightness, warmth, or spontaneity.
• A dark, bitter, high-strength mixture might generate slow, heavy motion with deep contrast, paired with a moodier, more introspective micro-story.

Both output forms—visual and text—are synchronized representations of the same underlying mood.

Technically, the project combines real-time camera input, segmentation, pre-trained recognition, and a generative graphics engine (p5.js, WebGL, or Three.js) with an AI-augmented story generator. The goal is not perfect recognition, but expressive mapping: helping users see how the ingredients and gestures of mixing shape an immediate aesthetic response. The final output can be exported as a still image or short generative video, becoming a unique record of each personal recipe.

Process

• Set a fixed camera position and defined glass area for consistent detection.
• Use CV to detect the glass and track the liquid surface for real-time volume estimation.