With image generation, Goodfellow et al. [1] defined the usage of Generative Adversarial Networks (GANs). With this, we can use two Neural Networks (NNs) that can help each other improve the model. For training, one of the NNs will be used to convert an image from pure noise into a recognisable image. In order to enhance the creativity of the images created, we can integrate Diffusion Models [2], and which adds extra noise from images in a number of steps. It is likely, soon, that we will not be able to tell real photographic images apart from ones that have been generated by GenAI.
With image generation, Goodfellow et al. [1] defined the usage of Generative Adversarial Networks (GANs). With this, we can use two Neural Networks (NNs) that can help each other improve the model. For training, one of the NNs will be used to convert an image from a purely noise into a recognisable image. In order to enhance the creativity of the images created, we can integrate Diffusion Models [2], and which adds extra noise from images in a number of steps. It is likely, soon, that we will not be able to tell real photographic images apart from ones that have been generated by GenAI.
Huggingface
The world of GenAI is now exploding, and you only have to look at the number of models on Huggingface to see that a whole new world of AI is being built:
