Terms included in the glossary

CFG Scale (Classifier-Free Guidance Scale) acts as a dial that controls how closely Stable Diffusion follows your written instructions (your prompt) when generating an image.

Click to read the full article

A checkpoint in Stable Diffusion is like a saved game file. Imagine the checkpoints used in race videogame.
It stores all the knowledge and skills the AI model has learned during its training process.

Click to read the full article

They perform two main jobs: first, they systematically add noise to an image until it's just fuzzy randomness.
Secondly, and crucially, they also learn how to reverse this process, starting with that noisy image and gradually removing the noise to reveal the image that a text prompt describes.

Click to read the full article

Fine-tuning, in machine learning, is the process of taking a model that has already been trained on a large, general dataset and adapting it to a smaller, more specific dataset.

Click to read the full article

LoRA (LOw-Rank Adaptation) are small upgrades you can add to your base model.
They teach your model a new skill, like how to draw a specific character (what you see the most), object or art style.

Click to read the full article

U-Net is a specific type of convolutional neural network architecture initially designed for biomedical image segmentation.
However, its effectiveness has led to its application in various image-to-image translation tasks.

Click to read the full article

A VAE (Variational Autoencoder) acts like a clever image compressor and decompressor within Stable Diffusion.
It takes a picture, squeezes it into a compact code, and then knows how to perfectly rebuild the original image from that code.

Click to read the full article