Pre-LLM Mainstream
GPT-2 / rinna and Deepfake were connected to scripts, live theater, and moving image. Generative AI was still handled as models, data, and implementation rather than a general chat interface.
Education / Generative AI / Storytelling
Data Driven Art 2020-2022
A record of Daito Manabe's course from 2020 through 2022, connecting machine learning and generative AI to coursework and student production. It spans the moment before LLMs became a mainstream social interface and continues into the era when GPT-derived tools, Deepfake workflows, VQGAN+CLIP, NFTs, Stable Diffusion, Midjourney, and ChatGPT entered creative practice.
Historical Position
The importance of this material lies in the fact that it preserves a concrete classroom and student-work record from just before generative AI became an everyday application to the point where its interfaces opened to general creative practice.
In 2020, the situation was not yet one in which anyone could use a conversational LLM in a browser. GPT-3 had started to change expectations around large language models, but the practical materials available in coursework were closer to GPT-2-based Japanese models such as rinna, DeepFaceLab-style Deepfake tools, live chat systems, and video editing pipelines.
In 2021, Colab-based text-to-image workflows such as VQGAN+CLIP made prompts, generated images, NFTs, and social-platform moderation part of the same artistic problem. Before image generation became a convenient production tool, the course treated it as a way to question value, ownership, national imagery, erotic imagery, and platform judgment.
In 2022, the public release of Stable Diffusion, the spread of Midjourney, and the arrival of ChatGPT at the end of the year moved generative AI from research and experimental workflows toward general creative environments. This page records how machine-learning-based art education framed that transition and turned tools into subject matter.
Prompt as Method
VQGAN+CLIP, NFTs, and Instagram moderation were treated together. Prompt-based image generation became a method for questioning markets, national representation, and platform decisions.
Public Generative AI
The course crossed into the period of Stable Diffusion, Midjourney, and ChatGPT, as generative AI moved from specialist experiments to tools accessible to general creators.
Course Record
The course asked how big data and artificial intelligence in an internet-native society could be connected to storytelling, moving image, image generation, markets, and instrument-based expression.
Period
3 Years
Course themes and student work from fiscal years 2020, 2021, and 2022.
Source
27 Slides
Exported from Google Slides as PDF and optimized into public WebP slide records.
Methods
AI / Data
Text generation, Deepfake, VQGAN+CLIP, NFTs, and Stable Diffusion.
Format
Course
Includes guest lectures and workshop-style group production.
Overview
Each year placed a newly socialized generative technology inside real production constraints, testing expression, ethics, markets, and viewing experience at the same time.
In 2020, the program used Deepfake and GPT-2 / rinna to examine generated, disrupted, and reconstructed storytelling through live chat, theater, lip-sync, and romance-driven narrative. Playwright Makoto Ueda and actor Tsuyoshi Muro joined as guest lecturers, and students developed short film works or proposed compositional techniques through workshops.
In 2021, the focus shifted to VQGAN+CLIP, covering text-to-image generation, conversion of generated images into 3D objects, NFT minting, and platform moderation on Instagram. Image generation was treated not only as material production, but as a problem spanning products, national imagery, erotic imagery, and platform judgment.
In 2022, the program focused on Stable Diffusion. Yuya Hanai from Rhizomatiks Research introduced past machine-learning projects and technical context around AI VTuber Eilan Mitsua, while the student works examined the value of images and the expansion of instrument-based expression.
Year Themes
The three annual themes are organized according to the source slide structure.
2020 / GPT-2(rinna), Deepfake and Storytelling
Deepfake and Storytelling
- Connected Deepfake, text generation, and live theater.
- Groups produced short film works or proposed compositional techniques.
- Guest lecturers: Makoto Ueda and Tsuyoshi Muro.
2021 / VQCLIP GAN and NFT
VQGAN+CLIP and NFTs
- Worked with image generation from text input.
- Minted AI-generated images as NFTs and speculated on market connection.
- Used products, national symbols, and erotic imagery as subjects.
2022 / Stable Diffusion
Stable Diffusion
- Focused on image generation using Stable Diffusion.
- Yuya Hanai introduced machine-learning projects and AI VTuber technology.
- Works addressed image value and expanded instrument expression.
Works
The works shown in the source slides are reorganized by intention and production method.
In-Pression
A work in which GPT-2-generated chat messages were sent into a YouTube Live theater performance and mixed with messages from human viewers.
- Turns live chat into material for script generation.
- AI and audience input encourage improvisation and script reconstruction.
- Video source: YouTube Live.
Lip-sync Conte
A prototype set in an everyday scene where another world exists inside a mirror, and the performer talks with another version of himself inside it.
- Synchronizes a dialogue video with the character's mouth movement.
- Generated the lip-sync video using DeepFaceLab.
Shojijo
A film work that expresses a world where Deepfake is used in daily life through a story of romance and self-transformation.
- Depicts a protagonist who impersonates the face preferred by someone they are interested in.
- Deals with anxiety and conflict around everyday Deepfake use.
Nonexistent Products
A work that generates product images from arbitrary product names, turns them into 3D objects, and makes them usable in a virtual reality space.
- Example prompt: “Perfect Design Pro Slim Vacuum Cleaner”.
- Assumes a near future where AI generates product descriptions and advertising.
- Questions whether AI-generated products are truly necessary inside capitalism.
Nations Seen Through Anthems
A video work that uses national anthem lyrics as VQGAN+CLIP input and watches images being generated in time with the anthem.
- Shows AI reading features from each country's lyrics.
- The source slides show outputs for the United States, Japan, and the United Kingdom.
x-porno
A work that tested how erotic images made with AI image generation would be evaluated today by posting them to Instagram.
- Some generated images were removed for violating community guidelines.
- The work frames a situation where AI-generated pornographic images are moderated by AI.
After Michael Mandiberg
A work that reflects on the future value of images in an era when anyone can generate and modify high-resolution images from text.
- Transforms an image into text, then that text back into an image.
- Continues transforming the resulting image into new images.
Breaking Instruments with Stable Diffusion
A work that defines an instrument as a device for intentionally producing sound through performance, then asks how Stable Diffusion can expand that definition.
- Connects not only hearing but also generated images to the performance experience.
- The blue body of a guitar transforms into water while the music changes with it.
Video
The YouTube Live link for In-Pression from the source slides is preserved here as an embedded video.
Source Slide Record
The original Google Slides, PDF, and PPTX are not placed inside the public route. All 27 pages are preserved as optimized display WebP files, including course overviews, diagrams, screenshots, and the QR-code slide.
Source Boundary
The raw Google Slides, PDF, and PPTX are kept outside the public Selected Works route. The public page contains only optimized display images, structured text, and the video embed.
Credits
People and guest information visible in the source slides.
- Presentation
- Daito Manabe
- 2020 Guest Lecturer
- Makoto Ueda
- 2020 Guest Lecturer
- Tsuyoshi Muro
- 2022 Guest Lecturer
- Yuya Hanai / Rhizomatiks Research
- Program Focus
- Data-driven art, generative AI, storytelling, image generation
AI Reference Surface
Related reference pages
Open the FAQ, glossary, authority, measurement, and AI index pages. Each link now states what it is for.