Link to this page location: #top
Link to this page location: #top

ai/ml in visuals

with vadim epstein

Link to this page location: #about

This author's course is targeted at practicing designers and media artists who work with modern digital imagery and wish to expand their creative palette to the current state of AI/ML progress.

We offer a practical and unique path to establish a solid ground for AI/ML experiments. The material represents a balanced combination of theory and custom open-source tools. The program cuts many corners of the usual presentation of AI/ML but does not trivialize the learning process, allowing you freedom in how you absorb the material.

Within 8 weeks, typically with two lessons per week, you will get the core theory of Visual AI/ML along with handy instruments – both locally installed and cloud-based – including GANs, Stable Diffusion, and other practical tools for image processing. After the fourth week, there is a spare week allocated for independent study, with the last week dedicated to your project.

Start

Autumn

Duration

8 weeks

Your skill level

Beginner, Intermediate

Software

PyTorch/TouchDesigner

Commitment

2 lessons per week

End date

Soon

Format

Online, Curated

Who needs this

Motion Designers Media Artists 3D Artists VR / XR Producers NFT Artists Interactive Developers Musicians VJs AI/ML Enthusiasts Designers

Streamlined complexity

The course can be seen as a comprehensive guided tour that explains the principles and details of Neural Networks operations without delving into complex mathematical formulas or advanced programming. The architectures and their specifics are covered up and down – from the big picture to the low-level experiments.

Clear perspective

One of the most important aspects of the course is the holistic Creative Coding approach, which connects ’classic’ generative practices like TouchDesigner to modern AI/ML applications. While it may lack some hot niche topics like prompt engineering, it provides a uniform understanding of the area and its place in the modern high-tech artistry.

Creative origin

The author of the course is a recognized practicing artist, therefore the program is built upon an artistic perspective, exploring the aesthetic (and/or even semantic) aspects of the technology alongside its functional features. The course also goes beyond popular toolkits and includes a few lesser-known tricks and techniques introduced by the author.
In short, it’s by an artist and for artists.

Applied outcomes

Alongside developing general concepts, students will master specific tools – from DeepDream to GANs to Stable Diffusion – set up at their disposal. The course won’t compete with the army of Youtube tutorials focused on tweaking UIs; instead, it complements them with a view from the inside.

Absolutely, if:

you are ready for intensive, active, self-directed work

you actively assimilate diverse materials

you're eager to communicate w/ teachers and students

you are seeking distinctive AI/ML mastery

Maybe not, if:

you are looking for effortless step-by-step instructions

you are frightened by uncertainty in the work

you expect the teacher to hold your hand

you are looking for an easy way to master AI/ML

Absolutely, if:

Maybe not, if:

you are ready for intensive, active, self-directed work

you actively assimilate diverse materials

you're eager to communicate w/ teachers and students

you are seeking distinctive AI/ML mastery

you are looking for effortless step-by-step instructions

you are frightened by uncertainty in the work

you expect the teacher to hold your hand

you are looking for an easy way to master AI/ML

Link to this page location: #galery

1.1. Intro to the course
Python, Colab vs local, libraries and modules. Framework setup.
1.2. Introduction to the neural networks
Neuron, graph, loss function. Training as an optimization process. CPPN (image reproduction)

2.1. CNN: concept, features
Convolutions as graphics filters, how they process & store data.
2.2. Architectures & building principles
Style transfer, autoencoders, GAN. Features, iterative pipeline, network extensions.

3.1. Modern GANs
ProGAN, StyleGANs. Latent space - animation, blending, projection, special tricks.
3.2. StyleGAN2 training
Data preparation, training process. Augmentation.

4.1. Image transformations (image-to-image)
From pix2pix to StarGAN2. Applied methods overview.
4.2. Integration with TouchDesigner
OSC, NDI, built-in Python with TouchEngine

5.1. Multimodal synthesis with foundation models
CLIP, early iterative methods. CPPNs, Aphantasia, VQGAN.
5.2. Transformers & Diffusion models
Attention. Denoised & latent diffusion.

6.1. Stable Diffusion as the current generative tool #1
Architecture, use cases: txt2img, img2img, interpolations, video. Controlnet
6.2. SD finetuning. Other tools & libraries
Textual inversion, Custom diffusion, LoRA. Diffusers, ComfyUI.

Development of the own project with visual AI/ML methods

Link to this page location: #team

Vadim Epstein

Course Author

Stanislav Glazov

Instructor for TouchDesigner integration

Vadim Epstein

MEDIA ARTIST, DIRECTOR, EDUCATOR, CODER, VJ

Former IT consultant and casual theoretical physicist, combining serious technical background, strong corporate experience and vivid creative mind. Has worked in various fields such as net.art and science art since 1996, eventually focused on visual media with stochastic algo narratives.

As an artist and curator, had made visuals for hundreds of concerts, festivals, parties, and commercial events. The artworks have been exhibited worldwide in Montreal, Vancouver, Stuttgart, Paris, London, Moscow, Lille; highlighted on the conferences NeurIPS 2020 / 2021 / 2022, CVPR 2021; sold as NFT collections, etc.

Besides commercial and personal projects, has delivered numerous talks, workshops and training courses.

More info

Link to this page location: #book

Minimum skills (heavily used throughout the course): general familiarity with command line (shell) operations and file paths. Get fluent with it in advance.
Programming skills will greatly help with exploring specific details of the used instruments.
Prerequisites are flexible: the more you know, the more you get.
The effectiveness of the course depends mostly on your learning skills and activity rather than on specific existing knowledge. Every student may have their own learning curve rather than following a fixed mandatory way.

Complete course program would require Windows computer with a decent Nvidia GPU with at least 6Gb VRAM (hard minimum is 1060, reasonable level 2070+).
Users with Macs would be able to use online Colab versions (no integration with TouchDesigner in this case).

Yes, a partial payment option is available: 50% before the start of the course and 50% one month after the start.

Yes, all practical topics have corresponding tasks to complete as homework. Most of them require you to repeat the actions from the lectures on your own. We recommend you execute them even if everything is clear in theory to ensure that you won’t face unexpected issues in your future work.
During the final week, you will have to develop your own project (with our assistance) as a kind of exam for the obtained knowledge.

10-12 hours per week should be sufficient to master the necessary concepts and skills. However, you may want to invest more time for more exhaustive explorations. Remember, the course program is open-ended: nearly all topics are not limited in scope and have lots of options for extra developments.

Link to this page location: #contacts