AI Design Glossary

An evaluation metric that measures how well a generated image aligns with the given text prompt. It quantifies prompt alignment by calculating the cosine similarity between text and image embeddings of the CLIP model.

Conditioning is the process of guiding an AI model to generate outputs based on specific inputs.

A neural network architecture that adds additional control layers to diffusion models to specify structural conditions like pose, edges, and depth maps during image generation.

A fine-tuning method developed by Google that customizes AI models to a specific subject or style using just a few photos.

The process of customizing a pre-trained AI model by providing additional training on a specific task, style, or dataset.

Knowledge distillation transfers knowledge from a large teacher model to a smaller student model.

A method for efficiently fine-tuning large AI models by adding small, trainable matrices. The original model weights remain unchanged.

Model merging combines weights of two or more models through mathematical methods to create a hybrid model.

Quantization reduces model size and speeds up inference by converting numerical values to lower bit formats.

An advanced prompt technique that assigns different text instructions to different regions of an image, providing independent content control for each area. It allows precise control over composition and layout.

The preservation of visual consistency between consecutive frames in video and animation generation. It encompasses the techniques used to ensure objects, characters, and backgrounds appear consistent from frame to frame.