AI Image Generation

Harsh Dalvi (202003004)

Bhanu Sunka (202003057)
Osama Shaikh (202003035)
Sarvesh Chavan (202003040)

Copyright @Information Technology Dept , XIE

 Introduction
• AI image generation has found applications in various domains, such as art, design,
entertainment, and even scientific research.

• Artists and designers are using these tools to explore novel visual styles, create digital
artwork, and generate new concepts.

• In entertainment, AI-generated images are used for special effects, game design, and
virtual world creation. Moreover, researchers are harnessing this technology for tasks
like data augmentation, image synthesis, and medical image generation.

• This project intersects technology, creativity, and innovation, promising a diverse scope
of applications across various industries.

AI Image Generation 2
 Objectives
• To create realistic and high-quality images. AI image generators can
be used to create images that are indistinguishable from real
photographs. This is useful for a variety of applications, such as
creating marketing materials, generating realistic product renders,
or creating digital art.

• To automate the image creation process. AI image generators can

automate the process of creating images, which can save time and
money.

• AI image generators can be used to create images that are beyond

the capabilities of human artists.

AI Image Generation 3
 Literature Review
Name Authors Year Presented at

Image Generation: a review Neural Elasri, M. O., Elharrouss, O., Research Gate 2022
Processing Letters Al-Maadeed, S., & Tairi, H.

V. Image Generation Based on Rohith, M., Pallavi, L., Shirisha, K., IEEE 2023
Text Using BERT And GAN Model. Sanjay, M., & Priya

Hierarchical Text-Condition Ramesh, A., Dhariwal, P., Nichol, IEEE 2022

Image Generation with CLIP A., Chu, C., & Chen, M.
Latents.

AI Illustrator: Art Illustration Zi-Han, C., Chen, L., Zhao, Z., & IEEE 2020
Generation Based on Generative Wang, Y.
Adversarial Network

Stacking VAE and GAN for Zhang, C., & Peng, Y. IEEE 2018
Context-aware Text-to-Image
Generation
 Name Authors Year Presented at

AraBERT and DF-GAN fusion for Bahani, M., Ouaazizi, A. E., & 2022 Science
Arabic text-to-image generation. Maalmi, K. Direct

MRP-GAN: Multi-resolution Qi, Z., Fan, C., Xu, L., Li, X., & Shu, Z 2021 Science
parallel generative adversarial Direct
networks for text-to-image
synthesis. Pattern Recognition
Letters

Image generation from text with Zhou, D., Sun, K., Hu, M., & He, Y. 2021 Science
entity information fusion. Direct
Knowledge Based Systems

DiverGAN: an efficient and Zhang, Z., & Schomaker, L. 2022 Science

effective Single-Stage framework Direct
for diverse Text-to-Image
generation.

AI Image Generation 5
 Literature Review
AI Illustrator: Art Illustration Generation Based on Generative
Adversarial Network

Zihan Chen, Lianghong Chen, Zhiyuan Zhao, Yue Wang With the
improvement of the spiritual need of the public, people have higher
requirements for books, among which the illustration of the relevant
words in the books is an urgent solution. The traditional method of
completing the related illustrations by illustrators has been unable to
meet the need of the growing book market.

AI Image Generation
6
Stacking VAE and GAN for Context-aware Text-to-Image Generation

Chenrui Zhang and Yuxin Peng As an attractive research topic, text-to-image

generation has been receiving extensive attention in computer vision and natural
language processing communities. This task aims to generate realistic images
conditioned on text description, which has widespread applications in various
fields, including photo editing and data augmentation, etc. Meanwhile, learning to
generate is a promising paradigm for semi-supervised and unsupervised learning,
which can provide meaningful hints for deep models’ interpretability.

AI Image Generation 7
Hierarchical Text-Conditional Image Generation with CLIP Latents

Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu Recent

progress in computer vision has been driven by scaling models on large
datasets of captioned images collected from the internet. Within this
framework, CLIP has emerged as a successful representation learner for
images. CLIP embeddings have a number of desirable properties: they
are robust to image distribution shift, have impressive zero-shot
capabilities, and have been fine-tuned to achieve state-of-the-art results
on a wide variety of vision and language tasks .

AI Image Generation
8
 Problem Definition

• Generating images that are realistic and coherent with the

provided input can be challenging.

• Handling of the diverse inputs and generate appropriate images.

• Balancing the trade-off between quality and speed is essential.

• Generating novel and original images that go beyond mere

replication of the training data is a critical challenge.

AI Image Generation
6
 References

[1]Elasri, M. O., Elharrouss, O., Al-Maadeed, S., & Tairi, H. (2022). Image Generation: a review.
Neural Processing Letters, 54(5), 4609–4646. https://doi.org/10.1007/s11063-022-10777-x

[2]Rohith, M., Pallavi, L., Shirisha, K., Sanjay, M., & Priya, V. (2023). Image Generation Based on
Text Using BERT And GAN Model. IEEE. https://doi.org/10.1109/cictn57981.2023.10141495

[3]Ramesh, A., Dhariwal, P., Nichol, A., Chu, C., & Chen, M. (2022). Hierarchical Text-Conditional
Image Generation with CLIP Latents. arXiv (Cornell University).
https://doi.org/10.48550/arxiv.2204.06125

[4]Zi-Han, C., Chen, L., Zhao, Z., & Wang, Y. (2020). AI Illustrator: Art Illustration Generation Based on
Generative Adversarial Network. IEEE. https://doi.org/10.1109/icivc50857.2020.9177494

[5]Zhang, C., & Peng, Y. (2018). Stacking VAE and GAN for Context-aware Text-to-Image Generation.
IEEE. https://doi.org/10.1109/bigmm.2018.8499439

AI Image Generation
7
 References
[6]Bahani, M., Ouaazizi, A. E., & Maalmi, K. (2022). AraBERT and DF-GAN fusion for Arabic
text-to-image generation. Array, 16, 100260. https://doi.org/10.1016/j.array.2022.100260

[7]Qi, Z., Fan, C., Xu, L., Li, X., & Shu, Z. (2021). MRP-GAN: Multi-resolution parallel generative
adversarial networks for text-to-image synthesis. Pattern Recognition Letters, 147, 1–7.
https://doi.org/10.1016/j.patrec.2021.02.020

[8]Zhou, D., Sun, K., Hu, M., & He, Y. (2021). Image generation from text with entity information fusion.
Knowledge Based Systems, 227, 107200. https://doi.org/10.1016/j.knosys.2021.107200

[9]Zhang, Z., & Schomaker, L. (2022). DiverGAN: an efficient and effective Single-Stage framework for
diverse Text-to-Image generation. Neurocomputing, 473, 182–198.
https://doi.org/10.1016/j.neucom.2021.12.005

AI Image Generation
8
Copyright @Information Technology Dept , XIE 9