Dream Collector: AI Interaction · Exploratory System
A web app and interactive installation that uses AI image generation to help people externalize fragmented dream memories as visual interpretations.
ROLE
Web front end · AI pipeline · TouchDesigner integration
TOOLS
OpenAI API · TouchDesigner · Web prototype
TEAM
Art-A-Hack™ 2025
TIMELINE
3 Days (June 2025)
Dream Collector: AI Interaction · Exploratory System
A web app and interactive installation that uses AI image generation to help people externalize fragmented dream memories as visual interpretations.
Web front end ·
AI pipeline · TouchDesigner integration
ROLE
3 Days (June 2025)
TIMELINE
Art-A-Hack™ 2025
TEAM
TOOLS
OpenAI API · TouchDesigner · Web prototype
OVERVIEW
OVERVIEW
Turning fragmented dreams into interactive visual interpretations
Turning fragmented dreams into interactive visual interpretations
Created as a selected team project for Art-A-Hack™ 2025 Special Edition at SVA New York, Dream Collector explores how AI can help people externalize scattered dream memories.
Created as a selected team project for Art-A-Hack™ 2025 Special Edition at SVA New York, Dream Collector explores how AI can help people externalize scattered dream memories.
Users enter short dream fragments into a web app, which generates three AI image interpretations. These images are then brought into a TouchDesigner installation, where visitors use hand gestures to reveal, distort, and navigate dream visuals in real time.
Users enter short dream fragments into a web app, which generates three AI image interpretations. These images are then brought into a TouchDesigner installation, where visitors use hand gestures to reveal, distort, and navigate dream visuals in real time.
Gesture interaction
Interactive painting
AI image generation
Dream memory
Gesture interaction
Interactive painting
CONCEPT
CONCEPT
Not one correct dream, but many possible readings
Not one correct dream, but many possible readings
The system was designed around interpretation rather than accuracy. By generating several visual directions from the same fragment, the experience gives users space to compare, question, and re-enter their own memory.
The system was designed around interpretation rather than accuracy. By generating several visual directions from the same fragment, the experience gives users space to compare, question, and re-enter their own memory.
The installation extends this private act into a shared visual field, where individual dreams become part of a living, layered archive.
The installation extends this private act into a shared visual field, where individual dreams become part of a living, layered archive.
Web prototype: users enter dream fragments and generate multiple visuals.
Installation: generated images become gesture-responsive dream visuals.
PROCESS
PROCESS
From prompt input to embodied visual exploration
From prompt input to embodied visual exploration
DAY 01
DAY 01
Testing the web flow
Testing the web flow
Started with a simple web interface to see how users would translate dream fragments into prompts.
Started with a simple web interface to see how users would translate dream fragments into prompts.
DAY 02
DAY 02
Building the image pipeline
Building the image pipeline
Built the front end and connected it to AI image generation through the OpenAI API.
Built the front end and connected it to AI image generation through the OpenAI API.
DAY 03
DAY 03
Connecting to TouchDesigner
Connecting to TouchDesigner
Linked the generated image pipeline into TouchDesigner to create an interactive visual environment.
Linked the generated image pipeline into TouchDesigner to create an interactive visual environment.
DAY 1
DAY 1
Framing the Problem
Framing the Problem
Early on, we explored multiple ways to translate dream input into visuals, including story-based generation, layered image composition, and gesture-driven navigation.
Through prototyping, we narrowed the system to short text input, multiple image outputs, and simple gesture-based exploration, prioritizing clarity and interaction over complexity.
Early on, we explored multiple ways to translate dream input into visuals, including story-based generation, layered image composition, and gesture-driven navigation.
Through prototyping, we narrowed the system to short text input, multiple image outputs, and simple gesture-based exploration, prioritizing clarity and interaction over complexity.
Web Prototyping
TouchDesigner Prototyping
DAY 2
DAY 2
Defining the Core Interaction
Defining the Core Interaction
We reduced web input to short phrases to lower the barrier to entry. Each prompt generates three images to support comparison and exploration.
The web app handles generation. The installation is reserved for interaction.
We reduced web input to short phrases to lower the barrier to entry. Each prompt generates three images to support comparison and exploration.
The web app handles generation. The installation is reserved for interaction.
Web Prototyping
TouchDesigner Prototyping
DAY 3
DAY 3
Designing the Visual System
Designing the Visual System
Visual outputs respond to emotional tone through subtle color shifts. We avoided rigid layouts to keep the experience ambient and reflective.
In the installation, distortion and gradual reveal mirror the unstable nature of dream memory.
Visual outputs respond to emotional tone through subtle color shifts. We avoided rigid layouts to keep the experience ambient and reflective.
In the installation, distortion and gradual reveal mirror the unstable nature of dream memory.
Web Prototyping
TouchDesigner Prototyping
DAY 4
DAY 4
Integrating the System
Integrating the System
We connected the web app and TouchDesigner through a shared data flow. This allowed AI-generated images to be reused and controlled in real time.
Gesture-based interaction replaced traditional UI controls to keep the installation non-verbal and immersive.
Given the timeline, we focused on a polished prototype rather than a public deployment.
We connected the web app and TouchDesigner through a shared data flow. This allowed AI-generated images to be reused and controlled in real time.
Gesture-based interaction replaced traditional UI controls to keep the installation non-verbal and immersive.
Given the timeline, we focused on a polished prototype rather than a public deployment.
Web Prototyping
TouchDesigner Prototyping
IMPLEMENTATION
IMPLEMENTATION
A lightweight pipeline between web generation and installation
A lightweight pipeline between web generation and installation
Web system
Web system
Dream fragment input
AI image generation via OpenAI API
Multiple visual outputs for one memory
Simple interface for fast exploration
Dream fragment input
AI image generation via OpenAI API
Multiple visual outputs for one memory
Simple interface for fast exploration
Installation system
Installation system
Generated images imported into TouchDesigner
Layered visual composition
Gesture-responsive reveal and distortion
Shared archive of personal dream visuals
Generated images imported into TouchDesigner
Layered visual composition
Gesture-responsive reveal and distortion
Shared archive of personal dream visuals
CHALLENGE
CHALLENGE
Designing for ambiguity without making it feel broken
Designing for ambiguity without making it feel broken
Dream descriptions are vague by nature. A single AI output can feel too definitive, while too many options can feel disconnected. The challenge was to keep the system open-ended without making the experience feel random.
We treated AI as a collaborator for interpretation, not as a tool for explaining what a dream “means.”
Dream descriptions are vague by nature. A single AI output can feel too definitive, while too many options can feel disconnected. The challenge was to keep the system open-ended without making the experience feel random.
We treated AI as a collaborator for interpretation, not as a tool for explaining what a dream “means.”
TAKEAWAYS
TAKEAWAYS