Text2Image_AI_Presentation

1.

Text2Image AI
Text prompt → Model → Image
PROMPT
MODEL
IMAGE
Presenter: Pilipenko Vladislav
Group: IU7-26B
Turning words into pictures

2.

Presentation Plan
1
What are Text-to-Image Models?
5
Diffusion Models
2
Where Are They Used?
6
Conclusion and Future
3
Main Types of Models
7
Questions
4
Autoencoders
2

3.

What Are Text-to-Image Models?
Example prompt
“A teddy bear on a skateboard
in Times Square”
Text-to-image models turn a
written prompt into a new
picture. The model reads the
words, understands the main
idea, and creates an image that
matches it. This helps people
make images without drawing
them by hand.
AI output image
3

4.

Where Are Text-to-Image Models Used?
Fashion
Advertising
Memes
These models are used in
fashion, advertising, and
online media. Clothing
companies can create new
design ideas faster. People
also use them for posters,
social media images, and
internet memes.
Fast ideas • low cost • many styles
Fashion design
Internet memes
4

5.

Main Types of Text-to-Image Models
There are three main groups: autoencoders, GANs, and diffusion models. They all create images,
but they do it in different ways. Today, diffusion models are the most popular for high-quality
results.
Autoencoders
VAE
VQ-VAE
GAN
older but
important
Diffusion
best image quality
today
Simple comparison of the main families
5

6.

Autoencoders
An autoencoder has two parts: an encoder and a decoder. The encoder compresses the image into a
small code, and the decoder builds the image again. This helps the model learn important features,
but some detail can be lost after compression.
x
input
encoder
decoder
e
d
e(x
small
)
code
d(e(x)
rebuilt
)
image
Example: details may be lost after strong
compression
6

7.

Diffusion Models
A diffusion model starts with
random noise. Step by step, it
removes noise until a clear
image appears. This process
is slower, but it can make very
detailed and realistic
pictures.
1
Start with noise
2
Predict what to remove
3
Repeat many times
Step
t
Noise
Predictor
(UNet)
Predicted
noise
Repeated denoising turns noise into an
image
7

8.

Conclusion and Future
Text-to-image models are powerful tools for art, design, and education. In the future, they will
become faster, safer, and easier to control. People will create better images in less time and with
more accuracy.
Better quality
More control
Safer use
Sharper and more realistic
images.
Users will guide style and
details more easily.
Rules and filters will
improve responsible use.
8

9.

Questions?
Thank you for your attention!
1. Which type of model is the most popular
today?
2. What is one real use of text-to-image
models?
I will be happy to answer.