Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Conditional Diffusion Model
Generative model architecture that incorporates conditional information (text, image, or other modalities) to control and guide the generation process toward specific results.
Noise Scheduler
Strategy that defines how noise is progressively added to data and removed during generation, directly influencing the quality and diversity of results.
Timestep Embedding
Numerical representation of time in the diffusion process that allows the model to know which denoising stage it is at and adapt its behavior accordingly.
Multi-Modal Diffusion
System capable of generating data based on multiple types of conditions simultaneously (text + image + audio), offering granular control over generation.
Text Inversion
Technique that learns a special text embedding to represent a new visual concept (style, object, person) without modifying the model's weights.
Depth Conditioning
Conditioning technique that uses depth maps to control the 3D structure and spatial arrangement of generated objects.
Semantic Segmentation Conditioning
Method that guides generation using semantic segmentation masks to precisely control the location and shape of different object categories.
Canny Edge Conditioning
Control approach that uses contours detected by the Canny algorithm as the basic structure for image generation.