23-02-2024
I have watched many YouTube videos yesterday to understand how AI image generation works. I came across quite a few new terms repeatedly and have noted them down on this page. I started watching videos about each term individually to further improve my understanding when watching tutorials on Comfy UI or other AI tools.
Here are some important terms specifically related to AI image generation within the Comfy UI.
- Model: In AI image generation, a model is a mathematical representation of a real-world process. It's a program that is trained on a dataset and can make predictions or decisions without being explicitly programmed to perform the task.
- Check Points: These are particular instances of a model saved during training, which can be used to continue training later from the same state, or to evaluate the model's performance.
- IP adapter: Image Prompt adapter. As the name suggest it is something that is used to use images as a prompt.
- Lora: LoRA stands for Low-Rank Adaptation. It's a training technique for fine-tuning Stable Diffusion models. LoRA is a popular method for fine-tuning large language models,diffusion models, and other AI models
- ControlNet: ControlNet is a neural network structure to control diffusion models by adding extra conditions.
- In simple words, Controlnet is used to preserve certain attribute of an image while altering others, example preserve colors from input image and change the scene, preserve pose from an image but change the person etc.
- PCA (Principal Component Analysis): This is a statistical procedure used in AI that converts possibly correlated variables into a set of values of linearly uncorrelated variables.
- Latent Space: In AI image generation, latent space refers to the abstract space in which the features of the images are captured. The AI model maps input data into this latent space and then decodes it back into the output image. This space is often multidimensional and each dimension represents a different feature or characteristic of the data.
- VAE - variational Auto encoders - Variational Autoencoders (VAEs) are a type of machine learning model used for generating new data that is similar to the training data. They're commonly used in tasks that involve the creation of new content, like image or text generation.
This page is giving me a really solid understaning of comfyui workflow
https://github.com/cubiq/ComfyUI_Workflows/blob/main/basic/README.md