admin

GANs

by

Generative Adversarial Networks are a type of neural network architecture invented in 2014 by Ian Goodfellow and his collaborators. GANs are foundational to much of today’s AI image generation.

A GAN is made of two neural networks that play a game:

ComponentRole
Generator (G)Tries to create fake data that looks like real data (e.g., fake images).
Discriminator (D)Tries to tell real from fake — it acts like a critic or detective.

They train together:

  1. The generator creates an image.
  2. The discriminator decides if it’s fake or real.
  3. Feedback from the discriminator helps the generator improve.
  4. Over time, the generator gets so good the discriminator can’t tell the difference.

This is why it’s called adversarial — the two networks are in a constant battle.

GANs are unsupervised (or self-supervised) learning models — they don’t need labeled data.

They learn the distribution of training data and generate new data from that distribution.

Many improved GANs have been developed since 2014, including:

VariantPurpose
DCGAN (2015)Deep Convolutional GAN — popular for image generation
StyleGAN (2018–2021)Introduced “style” control — used in “This Person Does Not Exist”
CycleGANImage-to-image translation (e.g., horses ↔ zebras)
BigGANHigh-quality, class-conditional image generation (from ImageNet)

StyleGAN

by

StyleGAN, developed by NVIDIA Research, is a groundbreaking architecture for generating ultra-realistic synthetic images, especially of human faces. Its ability to control “styles” across image layers set a new standard in AI image generation and led to viral real World applications like This Person Does Not Exist. Today, it’s used across art, games, fashion, and media, with both exciting and troubling implications.

The lead authors are Tero Karras, Samuli Laine and Timo Aila. Unlike earlier GANs, which often had limited control over image attributes, StyleGAN introduced a “style-based” architecture that revolutionized image synthesis. Images are generated in a multi-scale, layered way : High-level (pose, identity, …), mid-level (features like eye shape, …), low-level (color, texture, …). A latent input vector is transformed into an intermediate latent space (W space).

The following list presents the timescale of the launch of different StyleGAN versions :

🔹 StyleGAN1 (2018)
  • Introduced style-based generation
  • Produced realistic but occasionally distorted faces
🔹 StyleGAN2 (2019–2020)
  • Major quality improvement
  • Fixed artifacts and strange features in faces (e.g., weird teeth or asymmetry)
  • Used in “This Person Does Not Exist”
🔹 StyleGAN3 (2021)
  • Introduced equivariance, making it better at handling rotation and translation
  • Improved realism and temporal coherence (useful for video)

DALL-E

by

The first image generator created by OpenAI was called DALL-E and revealed in a blog post on 5 January 2021.

The name DALL·E is a creative blend of :

🎨 Salvador Dalí (the surrealist artist)

and

🤖 WALL·E (the animated robot from the 2008 Pixar movie)

🧠 It symbolizes :

  • Dalí → Represents creativity, surrealism, and imagination — aligning with the AI’s ability to generate fantastical and artistic images.
  • WALL·E → Represents technology, intelligence, and a lovable machine — highlighting the AI/robotic aspect of the model.

The AI model DALL-E was never made available to the public. DALL-E2 and DALL-E3 were the successors of DALL-E and they were both accessible by ordinary people.

In July 2021, Boris Dayma developed an open-source image generator called DALL-E Mini.

Crayon alias DALL-E mini

by

Boris Dayma initially created and launched DALL·E Mini during a hackathon organized by Hugging Face, in collaboration with Google in July 2021. This community event brought together around 800 participants, during which Boris Dayma built the first iteration of what would become a viral, open-source text-to-image model. This image generator was based on smaller, open-source AI models like VQGAN and CLIP. It was trained on publicly available datasets. It brought AI art to the masses long before full access to other image generators like DALL·E 2, Midjourney, or Stable Diffusion was common.

After it gained popularity, OpenAI reportedly asked them to rebrand to avoid misleading people that it was affiliated with, or endorsed by, OpenAI. The project was renamed Crayon, a simple and amicable rebranding, without a legal lawsuit.

DALL-E2

by

OpenAI officially unveiled the second version of it’s famous AI image generator DALL·E 2 on April 6, 2022, marking its debut to the public and the broader AI community.

📅 Key Launch Timeline:

  • April 6, 2022 – DALL·E 2 introduced, showcasing its significantly improved image quality and resolution .
  • July 20, 2022 – Entered beta with invites to approximately 1 million users
  • September 28, 2022 – Made publicly accessible by removing the waitlist

Among the first DALL-E2 images created by beta tester were the two following pictures published on July 22, 2022 :

Prompt : A Picasso-style painting of a parrot flipping pancakes ; Credit : IEEE Spectrum July 2022

Prompt : New Yorker-style cartoon of an unemployed panda realizing her job eating bamboo has been taken by a robot ; Credit : IEEE Spectrum July 2022

I was admitted as beta Tester by OpenAI on August 15, 2022. Here is my first image generated with the prompt cartoon of an elephant and a giraffe riding a bicycle on the beach. I posted it the same day on Facebook.

In the following days I generated more images and posted them as Image of the Day on Facebook. Here is a small gallery :

My first prompt became my favorite one to test new image generators when they became available. During your visit of posts about other images generators in the present knowledge section of the KI-Léierbud you will discover more images showing elephants and giraffes riding bicycles.

NodeJS apps

by

A minimal NextJS app consist of the following 2 files : 

my-next-app/
├── package.json
├── pages/
│   └── index.js

Here is a typical package.json file : 

{
  "name": "my-next-app",
  "version": "1.0.0",
  "scripts": {
    "dev": "next dev",
    "build": "next build",
    "start": "next start"
  },
  "dependencies": {
    "next": "latest",
    "react": "latest",
    "react-dom": "latest"
  }
}

It can be created manually or with npm init -y and then edited.

Here is a typical index.js file : 

export default function Home() {
  return <h1>Hello from Next.js on Plesk!</h1>;
}

To install the NextJS app in a Plesk server in a sub-domain we run the following commands in the Plesk terminal :

cd sub-domain.ki-leierbud.lu/my-next-app
npm install
npm run build

A lot of files and folders is created in the /my-next-app folder.

Development of complex NextJS apps

The AI tool lovable.dev is a great development assistent for creating powerful NextJS apps. Here is a list of a few wordpress articles which I edited to help coders to use this tool :

Lovable Applications

by

I created the following lovable.dev applications which should be compiled and hosted in the KI-Léierbud Plesk Server :

ki-leierbud-navigation

Link : https://ki-leierbud-navigation.lovable.app/

This app is used to access wordpress posts related to navigation (TOCs, …) in the KI-Léierbud portal with REST API.

The ID of a navigation wordpress post is appended with ?p=ID to the URL of the app.

Example : https://ki-leierbud-navigation.lovable.app/?p=12

ki-leierbud-knowledge

Link : https://ki-leierbud-knowledge.lovable.app/

This app is used to access wordpress posts related to knowledge in the KI-Léierbud portal with REST API.

The ID of the first wordpress post related to knowledge in a specific category is appended with ?p=ID to the URL of the app. Surfing through posts in the same category can de done by swiping or with forward and backward buttons. Language switching can be done with a language selector in the header of the webpage. Only the languages where a translation is available are shown in the language list.

When a link is clicked to another post in the portal with the current language settings ?lang=xx and the related post is not available in that language, the post with the next language in the priority list is fetched.

Examples :

To do :

  • add management to show author in the footer
  • define design with a palette of background colors
  • check the correct navigation
  • integrate the app into the subdomain knowledge.ki-leierbud.lu

ki-leierbud-quiz

Link : https://ki-leierbud-quiz.lovable.app/

This app is used to access wordpress posts related to quizzes in the KI-Léierbud portal with REST API.

The ID of the first wordpress post related to a specific quiz in a specific category is appended with ?p=ID to the URL of the app. Surfing through a quiz (posts in the same category) can de done by swiping or with forward and backward buttons. Language switching works similar as in the knowledge app.

A selection list with multiple answers is shown in the header of the quiz-page. The image or page number of a quiz is displayed in superposition on each page, the selected answer to a quiz-question is shown in superposition at the bottom of the page. The button to submit the quiz is only enabled if all the questions have been answered. After the submission a evaluation page with the results is returned to the user.

Examples of Image Style Quizzes :

To do :

  • select the final images
  • define design with a palette of background colors
  • check the correct navigation
  • integrate the app into the subdomain image-quiz.ki-leierbud.lu

ki-leierbud-polling

Links :

This example app is used to access wordpress posts related to pollings about emotions in the quiz section of the KI-Léierbud portal with REST API.

The ID of the first wordpress post related to a specific polling in a specific category is appended with ?p=ID to the URL of the app. Surfing through a polling (posts in the same category) can de done by swiping or with forward and backward buttons. Language switching works similar as in the knowledge app.

A selection list with multiple emotions is shown in the header of the quiz-page. The image or page number of a polling is displayed in superposition on each page, the selected answer to a polling-question is shown in superposition at the bottom of the page. The button to submit the polling is only enabled if all the questions have been answered.

The emotions selected by a user are stored inside a specific database created in the Plesk server. After the submission an evaluation page with the results is returned to the user showing a comparison with emotions selected by other users.

Examples of pollings :

To do :

  • select the final images and videos
  • select the final emotions
  • define design with a palette of background colors
  • check the correct navigation
  • integrate the app into the subdomains image-polling.ki-leierbud.lu and video-polling.ki-leierbud.lu.