Gary Boodhoo UX & Product Design

DeepDream Vision Quest, science fiction cave painting with machine learning

My Role

Creator & Developer

Product Type

Art Installation

DeepDream Vision Quest was an interactive generative AI art installation exhibited at festivals, galleries, and events from 2016–2019. After debuting at The Game Developers Conference, I became interested by how easily participants ascribed intention to the technology. Through a mix of outreach and word-of-mouth, it reached the San Francisco Palace of Fine Arts, the California Academy of Sciences, the Stanford Linear Accelerator, and other notable art and technology events.

The challenge was turning DeepDream from a slow generative process into a magic mirror people could step in front of and instantly grasp.

Partners

Lifecycle

Exhibition Period 2016–2019

Key Skills & Tools

In-house Game Tooling, Front-end Scripting, Photoshop, Illustrator, Prototyping

Credits

Creator & Developer Gary Boodhoo

DeepDream Vision Quest, science fiction cave painting with machine learning

My Role

Creator & Developer

Product Type

Art Installation

DeepDream Vision Quest was an interactive generative AI art installation exhibited at festivals, galleries, and events from 2016–2019. After debuting at The Game Developers Conference, I became interested by how easily participants ascribed intention to the technology. Through a mix of outreach and word-of-mouth, it reached the San Francisco Palace of Fine Arts, the California Academy of Sciences, the Stanford Linear Accelerator, and other notable art and technology events.

The challenge was turning DeepDream from a slow generative process into a magic mirror people could step in front of and instantly grasp.

Partners

Lifecycle

Exhibition Period 2016–2019

Key Skills & Tools

In-house Game Tooling, Front-end Scripting, Photoshop, Illustrator, Prototyping

Credits

Creator & Developer Gary Boodhoo

DeepDream Vision Quest, science fiction cave painting with machine learning

My Role

Creator & Developer

Product Type

Art Installation

DeepDream Vision Quest was an interactive generative AI art installation exhibited at festivals, galleries, and events from 2016–2019. After debuting at The Game Developers Conference, I became interested by how easily participants ascribed intention to the technology. Through a mix of outreach and word-of-mouth, it reached the San Francisco Palace of Fine Arts, the California Academy of Sciences, the Stanford Linear Accelerator, and other notable art and technology events.

The challenge was turning DeepDream from a slow generative process into a magic mirror people could step in front of and instantly grasp.

Partners

Lifecycle

Exhibition Period 2016–2019

Key Skills & Tools

In-house Game Tooling, Front-end Scripting, Photoshop, Illustrator, Prototyping

Credits

Creator & Developer Gary Boodhoo

DeepDream Vision Quest, science fiction cave painting with machine learning

My Role

Creator & Developer

Product Type

Art Installation

DeepDream Vision Quest was an interactive generative AI art installation exhibited at festivals, galleries, and events from 2016–2019. After debuting at The Game Developers Conference, I became interested by how easily participants ascribed intention to the technology. Through a mix of outreach and word-of-mouth, it reached the San Francisco Palace of Fine Arts, the California Academy of Sciences, the Stanford Linear Accelerator, and other notable art and technology events.

The challenge was turning DeepDream from a slow generative process into a magic mirror people could step in front of and instantly grasp.

Partners

Lifecycle

Exhibition Period 2016–2019

Key Skills & Tools

In-house Game Tooling, Front-end Scripting, Photoshop, Illustrator, Prototyping

Credits

Creator & Developer Gary Boodhoo

Interacting with machine learning in public, generative art at GDC

Speaker at The Game Developers Conference

Seeing Yourself Seeing Yourself, a circular buffer for live video experiments

Creative R&D · Skinjester Studio

DeepDream Vision Quest, codebase

Github repository

Process

Contemporary mechanical hallucinations

Process

Contemporary mechanical hallucinations

Process

Contemporary mechanical hallucinations

Process

Contemporary mechanical hallucinations

From the start there was a mismatch. Live video runs at 30 frames per second, while DeepDream hallucinations emerge slowly. A completed image might take 20 seconds to fully resolve, so I didn't wait for them to do so. It was more important to watch the dream emerge in real time than to see it perfectly sharp right away.

From the start there was a mismatch. Live video runs at 30 frames per second, while DeepDream hallucinations emerge slowly. A completed image might take 20 seconds to fully resolve, so I didn't wait for them to do so. It was more important to watch the dream emerge in real time than to see it perfectly sharp right away.

From the start there was a mismatch. Live video runs at 30 frames per second, while DeepDream hallucinations emerge slowly. A completed image might take 20 seconds to fully resolve, so I didn't wait for them to do so. It was more important to watch the dream emerge in real time than to see it perfectly sharp right away.

From the start there was a mismatch. Live video runs at 30 frames per second, while DeepDream hallucinations emerge slowly. A completed image might take 20 seconds to fully resolve, so I didn't wait for them to do so. It was more important to watch the dream emerge in real time than to see it perfectly sharp right away.

Multi-scale image processing with DeepDream
DeepDream operates in successive stages, each yielding a sharper, higher-resolution output. Showing intermediate steps kept the system visually responsive.

Reference image Stackademic via Medium

Multi-scale image processing with DeepDream
DeepDream operates in successive stages, each yielding a sharper, higher-resolution output. Showing intermediate steps kept the system visually responsive.

Reference image Stackademic via Medium

Multi-scale image processing with DeepDream
DeepDream operates in successive stages, each yielding a sharper, higher-resolution output. Showing intermediate steps kept the system visually responsive.

Reference image Stackademic via Medium

Multi-scale image processing with DeepDream
DeepDream operates in successive stages, each yielding a sharper, higher-resolution output. Showing intermediate steps kept the system visually responsive.

Reference image Stackademic via Medium

A game of hide-and-seek
Stillness triggers dreaming, motion returns to live video. Refreshing the screen with partial dream updates kept the interaction feeling responsive.

Runtime Video Development

A game of hide-and-seek
Stillness triggers dreaming, motion returns to live video. Refreshing the screen with partial dream updates kept the interaction feeling responsive.

Runtime Video Development

A game of hide-and-seek
Stillness triggers dreaming, motion returns to live video. Refreshing the screen with partial dream updates kept the interaction feeling responsive.

Runtime Video Development

A game of hide-and-seek
Stillness triggers dreaming, motion returns to live video. Refreshing the screen with partial dream updates kept the interaction feeling responsive.

Runtime Video Development

Hardware

The physical pipeline

Hardware

The physical pipeline

Hardware

The physical pipeline

Hardware

The physical pipeline

For every exhibition, my hardware rig was disassembled, transported, and reassembled onsite. Then broken down, packed, and stored again. Interesting how this physical pipeline mirrored the image generation pipeline. Over time, my logistics became rigorously simple, with labeled components tracked at every stage, including:For every exhibition, my hardware rig was disassembled, transported, and reassembled onsite. Then broken down, packed, and stored again. Interesting how this physical pipeline mirrored the image generation pipeline. Over time, my logistics became rigorously simple, with labeled components tracked at every stage, including:

  • Small form factor Core i7 workstation 16GB

  • NVIDIA Titan X (Pascal) GPU

  • 30" 4K monitor

  • Logitech C920 webcam (2)

  • 65" HDTV

  • Mobile flat panel AV cart

  • Mobile computer cart

  • TV backlighting

  • USB control keypad

  • C-stands and mounts

  • Mobile computer cart

  • Venue-specific lighting

  • Cabling

  • Power adaptors and strips

DeepDream Vision Quest installation rig

Photograph Gary Boodhoo

DeepDream Vision Quest installation rig

Photograph Gary Boodhoo

DeepDream Vision Quest installation rig

Photograph Gary Boodhoo

DeepDream Vision Quest installation rig

Photograph Gary Boodhoo

Software

Hide and seek

Software

Hide and seek

Software

Hide and seek

Software

Hide and seek

As I mentioned, the system dreamed only when the video feed was nearly still. I counted differences between frames and when this value fell below a threshold, the system began hallucinating. When motion picked up again, the live feed smoothly faded back in. The experience was a game of hide-and-seek. Transitioning smoothly between the live and dreaming states became the key driver of my development efforts.

As I mentioned, the system dreamed only when the video feed was nearly still. I counted differences between frames and when this value fell below a threshold, the system began hallucinating. When motion picked up again, the live feed smoothly faded back in. The experience was a game of hide-and-seek. Transitioning smoothly between the live and dreaming states became the key driver of my development efforts.

As I mentioned, the system dreamed only when the video feed was nearly still. I counted differences between frames and when this value fell below a threshold, the system began hallucinating. When motion picked up again, the live feed smoothly faded back in. The experience was a game of hide-and-seek. Transitioning smoothly between the live and dreaming states became the key driver of my development efforts.

As I mentioned, the system dreamed only when the video feed was nearly still. I counted differences between frames and when this value fell below a threshold, the system began hallucinating. When motion picked up again, the live feed smoothly faded back in. The experience was a game of hide-and-seek. Transitioning smoothly between the live and dreaming states became the key driver of my development efforts.

Motion detection monitor screen
This debug view visualized the motion-detection frame buffer and became an audience-friendly “how it works” display that sparked conversations.

Runtime Video Live Event

Motion detection monitor screen
This debug view visualized the motion-detection frame buffer and became an audience-friendly “how it works” display that sparked conversations.

Runtime Video Live Event

Motion detection monitor screen
This debug view visualized the motion-detection frame buffer and became an audience-friendly “how it works” display that sparked conversations.

Runtime Video Live Event

Motion detection monitor screen
This debug view visualized the motion-detection frame buffer and became an audience-friendly “how it works” display that sparked conversations.

Runtime Video Live Event

Motion detection telemetry for tuning
I logged frame difference pixel counts over time in Python and plotted them to see how parameter changes affected motion classification. This made tuning measurable, optimizing for responsive switching without false triggers.

Data Visualization Gary Boodhoo, Skinjester Studio

Motion detection telemetry for tuning
I logged frame difference pixel counts over time in Python and plotted them to see how parameter changes affected motion classification. This made tuning measurable, optimizing for responsive switching without false triggers.

Data Visualization Gary Boodhoo, Skinjester Studio

Motion detection telemetry for tuning
I logged frame difference pixel counts over time in Python and plotted them to see how parameter changes affected motion classification. This made tuning measurable, optimizing for responsive switching without false triggers.

Data Visualization Gary Boodhoo, Skinjester Studio

Motion detection telemetry for tuning
I logged frame difference pixel counts over time in Python and plotted them to see how parameter changes affected motion classification. This made tuning measurable, optimizing for responsive switching without false triggers.

Data Visualization Gary Boodhoo, Skinjester Studio

Events

Exhibition history

Events

Exhibition history

Events

Exhibition history

Events

Exhibition history

2019

Beyond Human Nightlife Event, California Academy of Sciences, San Francisco, CA, August 23

2018

The 5th Last Festival, Stanford Linear Accelerator, Stanford, CA, April 26–29

If So, What? (ISW) Festival, Palace of Fine Arts, San Francisco, CA, April 26–29

Lost Chord Awards for the Ritual, Sacred and Folk Arts, Qal’bu Maryam Women’s Mosque, Berkeley, CA, June 2

CODAME Art + Tech Festival, The Midway, San Francisco, CA, June 6–7

Noise Floor 2, Red Victorian, San Francisco, CA, August 25

2017

The 4th Last Festival, Hammer Theater, San Jose, CA, April 7–8

living.room, Soma Grand, San Francisco, CA, October 7

Noise Floor 1, The East Cut, San Francisco, CA, December 9

2016

CODAME Art + Tech Festival, Hotel Zetta, San Francisco, CA, November 11

Living Room, Soma Grand, San Francisco

Runtime Video Live Event

Living Room, Soma Grand, San Francisco

Runtime Video Live Event

Living Room, Soma Grand, San Francisco

Runtime Video Live Event

Living Room, Soma Grand, San Francisco

Runtime Video Live Event

CODAME Art + Tech Festival, Hotel Zetta, San Francisco

Runtime Video Live Event

CODAME Art + Tech Festival, Hotel Zetta, San Francisco

Runtime Video Live Event

CODAME Art + Tech Festival, Hotel Zetta, San Francisco

Runtime Video Live Event

CODAME Art + Tech Festival, Hotel Zetta, San Francisco

Runtime Video Live Event

DeepDream Vision Quest projected into a domestic interior, turning an everyday kitchen into an immersive, responsive visual environment.

Runtime Video Development

DeepDream Vision Quest projected into a domestic interior, turning an everyday kitchen into an immersive, responsive visual environment.

Runtime Video Development

DeepDream Vision Quest projected into a domestic interior, turning an everyday kitchen into an immersive, responsive visual environment.

Runtime Video Development

DeepDream Vision Quest projected into a domestic interior, turning an everyday kitchen into an immersive, responsive visual environment.

Runtime Video Development

Noise Floor, The East Cut, San Francisco

Runtime Video Live Event

Noise Floor, The East Cut, San Francisco

Runtime Video Live Event

Noise Floor, The East Cut, San Francisco

Runtime Video Live Event

Noise Floor, The East Cut, San Francisco

Runtime Video Live Event

CODAME Art + Tech Festival, The Midway, San Francisco

Photograph Gary Boodhoo

CODAME Art + Tech Festival, The Midway, San Francisco

Photograph Gary Boodhoo

CODAME Art + Tech Festival, The Midway, San Francisco

Photograph Gary Boodhoo

CODAME Art + Tech Festival, The Midway, San Francisco

Photograph Gary Boodhoo

Beyond Human Nightlife Event, California Academy of Sciences, San Francisco

Photograph Gary Boodhoo

Beyond Human Nightlife Event, California Academy of Sciences, San Francisco

Photograph Gary Boodhoo

Beyond Human Nightlife Event, California Academy of Sciences, San Francisco

Photograph Gary Boodhoo

Beyond Human Nightlife Event, California Academy of Sciences, San Francisco

Photograph Gary Boodhoo

Results

This project’s success opened doors to new professional opportunities, including a profitable creative AI workshop series and invitations to speak on prominent stages.

Partners

Lifecycle

Exhibition Period 2017–2019

Key Skills & Tools

In-house Game Tooling, Front-end Scripting, Photoshop, Illustrator, Prototyping

Credits

Creator & Developer Gary Boodhoo

Gary Boodhoo

Reliable systems. Creative execution.

Available for senior UX and product design roles, as well as contract work. I help teams build understanding and trust at key moments.

Resume

LinkedIn

48

Designed with Framer in SF by Gary Boodhoo

Last updated Mar 2026

v4.0

96