AI 용어집
인공지능 완전 사전
Text-to-Image Diffusion
Subcategory of conditional diffusion models where the input is a text description serving as a condition to generate a corresponding image.
Image-to-Image Diffusion
Conditional diffusion process where an input image is used as a guide to transform or stylize an output image, while preserving certain characteristics of the original.
Outpainting by Diffusion
Extension of an image beyond its original boundaries using a conditional diffusion model, which generates new pixels coherent with the existing content.
Conditional Timestep
Parameter in the diffusion process that can be modulated by an external condition to control the level of detail or style at a specific denoising step.
Guided Resampling
Method that iteratively adjusts the noise sample during the denoising process to ensure it remains aligned with the provided condition, strengthening control over generation.
Null Conditioning
Training strategy where the model learns to generate images without condition, used in parallel with conditioning to enable fine control over the level of adherence to the condition via guidance.
Prompt Engineering for Diffusion
Art and science of designing optimal conditioning texts (prompts) to obtain the most accurate and creative results possible from a text-to-image diffusion model.
Conditional Latent Diffusion
Approach that applies the diffusion process in a compressed latent space, conditioned by external information, for faster generation and more efficient resource usage.