Core Architecture & Processes
Algorithmic process where a model creates new visual data learned databases.
Image generation
AI makes images without directions, lets it create at random.
Unconditional
The process of selecting, filtering, labeling, and organizing training data for high quality coverage.
Dataset Curation
Techniques used during dataset preparation, training, or sampling to reduce behaviors in generated outputs.
Bias Mitigation
Process of predicting and subtracting noise to convert a corrupted image into a clean image.
Denoising
User enters a prompt to create a image
Conditional
The room number of this room
806
Methods to make sure a specific visual subject or character maintains recognizable identity traits across multiple generated images.
Character Consistency
Framework that creates data by a process that turns images into gaussian noise.
Diffusion model
Text to Image Generation
An image mask where each pixel is labeled according to the object class or region it belongs to, giving models clear spatial guidance.
Segmentation Map
Designed by Erwin Schrödinger to highlight the weirdness of quantum mechanics, this thought experiment features a hypothetical pet that is simultaneously alive and dead until someone opens a sealed box to observe it.
Schrödinger's cat
Deep learning model that looks at images by scanning them in small grids. Perfect for spotting shapes and textures.
Convolutional Neural Network ( CNN )
Taking a existing photo and changing its style.
Image to Image translation
The best CC we had
WP
The degree to which a machine learning model converges during training without issues.
Training Stability
Blue print that made diffusion models efficient at clearing noise
What did Areean Khan put in Neel's mouth on the first day of junior year
Neel's Glasses
A method that maps inputs into low dimensional mathematical vector spaces so the model can process their meaning.
Embedding-Based Approach
Introducing controlled noise during processing or generation to add variation.
Noise Injection