🖥️Technical explanations

How does our generative model work?

Our generative model is based on the Stable Diffusion method, which is a stochastic process based on the theory of partial differential equations. This method permits to generate images from a random noise by applying a series of transformations and convolutions. The process is iterative and uses transport equations to diffuse the random noise through a continuous space, while preserving key image properties such as regularity and continuity.

We trained our generative model using a public dataset, which we had to clean and preprocess to ensure the quality of the generated images. Then we used a stochastic gradient optimization algorithm to optimize the model parameters, minimizing a cost function that measures the difference between the generated images and the real images.

This training process was particularly expensive in terms of time and resources, requiring an investment of over 5,000$ to rent several high-performance computing machines and the storage space needed to process the data. The results were impressive, with a generative model that is capable of generating high quality images and can be used in a variety of applications.

Why are we different from others?

What makes us different from others is that most generative model users simply use pre-made models available online. But as for us, we decided to train our own generative model according to our own expectations and needs.

We invested time, money and resources to train an optimized and comprehensive generative model that can generate realistic and high quality images in a range of areas. We ensured that our model was specifically adapted to our needs, rather than simply using generic models that might not be optimal for our use.

Ultimately, our approach paid off, as we were able to develop a best-in-class generative model capable of generating portraits, photo realistic NSFW images, landscapes and more! This customized approach also allows us to better understand how our model works and to improve it even more in the future.