Good to have you here.
About Me
Projects
Music / joe
Connect
Fun Stuff
ASCII Cam
I'm Joseph Croskey, though most people call me Joe. I'm from Detroit, and for as long as I can remember, I've been drawn to making things. Music was one of my first obsessions. I started making it when I was young, and that early fascination with sound opened the door to a much bigger world of creativity. Around the same time, I was also starting little businesses in middle school, always looking for ways to turn ideas into something real.
Since then, that same curiosity has shaped the way I move through life. I've always been passionate about art, food, technology, storytelling, and exploring the world. I do not see myself as just one thing. More than anything, I'm someone who loves learning, creating, and connecting ideas across different mediums, cultures, and experiences.
That mindset has taken me across Detroit, Shanghai, London, and beyond. I studied Interactive Media and Business at NYU Shanghai, where I developed a global perspective and deepened my interest in how creativity and technology can work together. Later, I earned a Master's degree in Music Production with Distinction from the University of the Arts London, which further shaped the way I think about sound, design, and emotional impact.
Today, I work at Intelligent Internet, where my role has grown across marketing, creative, content, growth, and cross-functional communication. Over the past year, I've helped grow the company's social media presence by more than 100 percent, contributed to editing and publishing a number one Amazon bestseller, created launch videos that reached millions of views, and produced podcasts with hundreds of thousands of views. I've also worked across teams in China, Canada, Vietnam, and the UK, and being fluent in Chinese has allowed me to build genuine connections across markets and cultures.
At the core of everything I do is the same thing that has been there from the beginning: curiosity. Whether I'm making music, shaping a story, building a brand, designing something visual, or discovering a new place through its food and culture, I'm most fulfilled when I'm creating something meaningful.
Education
Bachelor of Science, Interactive Media + Business · Minor in Chinese
Master of Arts, Music Production · Distinction
Career
Intelligent Internet (ii.inc)
I help bring frontier open-source AI to the world through story, design, and launch. Working across creative, content, product communication, and brand, I turn complex ideas into media, experiences, and narratives that make emerging technology feel tangible, relevant, and impossible to ignore.
Freelance
Designed digital assets for artists and marketing teams. Created visual content for music releases, tours, and promotional events using Adobe Creative Suite.
Freelance
Over 1M+ total streams. Proficient in ProTools, Ableton, Logic Pro X, FL Studio. Crafted spatial audio for VR/AR projects and interactive art installations with surround sound 5.1 mixing.
Freelance
Played a pivotal role in shaping the hyperpop sound. Discovered and developed artists who achieved commercial success. Facilitated cross-cultural collaborations between Western and Chinese artists.
Fantasy Market · NYU Shanghai
Created NYU Shanghai's first collaboration with a local business. Partnered with New Bund 31 to display student art. Garnered local and international media coverage.
Freelance
Built accounts to hundreds of thousands of followers and successfully monetized them. Developed growth strategies including trend analysis, hashtag research, and engagement tactics.
Daimler AG (Mercedes-Benz)
Assisted in executive-level projects. Created presentations for the North American Operations Committee. Managed executive schedules and strategic plans.
"I like creating beautiful things"
2026 · Video, Sound Design, Coloring, Editing
Launch video for LOGOS, Intelligent Internet's frontier augmented intelligence engine. Shaped end-to-end — from concept through final frame, sound, and color.
2022 · AI + Magenta + BigGan
Collaborative project combining two AI applications to create audio-visual content. Melodies generated with the Max for Live device Magenta; visuals created using images from BigGan with interpolation between classes. Mix, master, and production by Jœ.
2022 · Midjourney + DiscoDiffusion
Exploring how machine learning expands the realm of art curation and creation. Features AI-generated cityscapes, installations, and portraits.
From teaching a program how to play checkers in the 1950s to a computer defeating a world class chess champion in 1997, it is needless to say that artificial intelligence is developing at a rate faster than anyone would have ever expected it to. While these applications of machine learning have the potential to make our lives easier, they also open up a whole other realm in the world of art curation and creation. The more advancements that we have in creating artificially "intelligent" programs, means the more possible applications they have for creative purposes. On top of that, the more data that these applications receive make them smarter over time, so the amount of possibilities for applications is endless. In this paper I will be discussing some recent advancements in artificial intelligence (GANs, CANs, and OpenAI's DALL·E) and how they are being used for creative purposes.
One of the most influential recent advancements in regards to using AI to generate art are called Generative Adversarial Networks or (GANs). Introduced in 2014, GANs significantly sped up and developed the process for using machine learning to create novel outputs. The main reason why these GANs were so influential was because they were the first generative model that gave satisfying results and they also paved the way for the basis of more research in regards to the world of machine learning. GAN research director Yann LeCun even said that GANs were "the most interesting idea in the last 10 years in ML". GANs are described as "algorithmic architectures that use two neural networks, pitting one against the other (thus the 'adversarial') in order to generate new, synthetic instances of data that can pass for real data." More specifically within the two networks, there is one generator that is responsible for generating outputs, and there is a discriminator that is responsible for evaluating the data. The discriminator also checks the generated data to see if it came from the training set or not. By going through these processes, GANs are able to create images that look real, but in reality they did not even exist beforehand and are completely new. Some examples of this include, generating faces that don't exist, bedrooms, anime characters, pokemon, and even image to image translation — the possibilities are endless. One of my personal favorite applications of GANs comes from a music video by Andrew Paley where he syncs GAN generated images to the audio. Paley did this by creating a self-developed GAN powered program called Pixie that syncs the music imported to the GAN generated images to create a fluid flowing video transition between the images that also correlates perfectly with the beat and tempo of the song. This type of artistic GAN application is pretty common nowadays and with this more and more people are finding ways to improve and expand the capabilities of GANs in regards to creating collaborative audio-visual projects.
Another advancement in machine learning (surprisingly enough that was branched out directly from GANs) goes by the name CANs or Creative Adversarial Networks. Because some people believed that GANs lacked some of the creativity needed in order to create entirely new and novel outputs, programmers proposed the idea of CANs in order to compensate. CANs stem from an idea by D.E. Berlyne that "the psychophysical concept of 'arousal' has a great relevance for studying aesthetic phenomena" (Elgammal, 2017), which can also be referred to as the Martindale principle. CANs aims to raise the arousal potential of machine generated images by modifying GANs and increasing their "stylistic ambiguity" output. Another way that CANs differ from GANs is that CANs discriminator is trained over a very large dataset of art and the generator will only create images from the ones that are deemed as art from the discriminator. These improvements to GANs are valuable because CANs are attempting to create a more artistic approach to machine learning, something that is incredibly valuable for artists and creators of things. Furthermore, the development of CANs can also serve as a building block for more programs and models that are specifically designed to create new never before seen art works. In my opinion I think that the function of GANs can definitely be useful for the creation of art and novel outputs, but I do agree with the opinion that GANs are probably not the best programs for creating art. But by tweaking the GAN model slightly and developing a new model (CANs), it's probable that this model is better suited to create more artistic outputs in the long term.
Moving forward, one of the largest and most significant recent advancements in AI in regards to creating novel outputs is OpenAI's 12 billion parameter text to image network. This network is called DALL·E and it works by using a large data set of text image pairs in order to generate images based upon text. On the OpenAI website they describe DALL·E's capabilities as: "creating anthropomorphized versions of animals and objects, combining unrelated concepts in plausible ways, rendering text, and applying transformations to existing images." After taking a look at some of the images it produced, I was quite surprised to say the least. In the past, there was a large wave of programmers attempting to create text to image programs like DALL·E, but they all ultimately ended up not working as wanted up until the introduction of Contrastive Language Image Pre Training (CLIP). CLIP was a game changer in creating a successful text to image program because it was a model that was trained upon 400 million text-image pairs that were collected from the internet. The DALL·E model relies on the CLIP model to rank the results of generated images in relation to the input text. This CLIP model is also being used recently to create more open source implementations like BigSleep and DeepDaze. With this advancement I imagine that within the next couple of years there will be even more breakthroughs in regards to creating novel outputs based upon a prompt of text.
Looking at these three applications of machine learning in regards to creating art and novel outputs, it's interesting to see how each of them overlap in some ways but ultimately differ in their outputs. From the first and extremely vital advancement of the GAN model, programmers were finally able to see satisfying results. Taking the GAN model further, researchers and programmers through the CAN model looked to create something that was better suited for creating more artistic outputs. Moreover, the revolutionary development of OpenAI's DALL·E that uses Contrastive Language Image Pre Training is a critical advancement because it was the first of its kind to create unbelievably accurate text-to-image outputs. Lastly, after researching and writing about these crucial advancements that each serve a vital role in the improvement and development of producing more intelligent machine learning programs, I am excited to see what the future holds in regards to AI and its relationship to art.
2022 · Canon 6D + Premiere Pro
Juxtaposition shot on Canon 6D in Shanghai. Edited and color graded in Premiere Pro. Original soundtrack by Jœ.
2021 · Processing
Generative circle patterns created using Processing. Recursive visual explorations of geometry and motion.
Visual Metaphor · Collaboration with Charlie Howes
A visual metaphor project created in collaboration with Charlie Howes. Controlled chaos — beautiful disorder as creative expression.
2022 · Photoshop + Illustrator · 2540×3548
Diptych, shot in Shanghai 2022. Commentary on privacy, technology, and the omnipresent lens.
2022 · Premiere Pro · Sounds by Jœ & Sovietaxi
Shot in Shanghai 2022. Video exploring the rhythm and ritual of navigating the city's metro system. Edited in Premiere Pro with an original soundtrack.
Spring 2022 · Illustrator
Advertisements created for placement in the NYC Metro system. Designed in Illustrator for transit advertising formats.
InDesign · 8 pages + covers
Mock Gentlewoman Magazine redesign and layout, playing around with image placement and type.
7+ years · Millions of plays
The project and artist Jœ, which came into existence in 2022, is the culmination of over 11 years of music creation, collaboration, and refinement — dating back to 2014.
The goal: to create a unique auditory universe — entirely self-produced, mixed, and mastered. With 1M+ total streams across platforms, the project spans ambient soundscapes, beat-driven explorations, and vocal-forward pieces at the intersection of music and art.
Proficient in ProTools, Ableton, Logic Pro X, and FL Studio. Specialized in spatial audio mixing for VR/AR projects and interactive media art installations using surround sound 5.1.
As an A&R, played a pivotal role in shaping the hyperpop sound — discovering and developing artists who achieved commercial success, and facilitating cross-cultural collaborations between Western and Chinese artists.
Trained at the University of the Arts London with a Master's in Music Production (Distinction).
SoundCloud
Open to conversations about creative projects, AI, web3, and making cool things.
Send a Message
> INITIALIZING CAMERA... > PLEASE ALLOW ACCESS
