AI-ordlista
Den kompletta ordlistan över AI
Diffusion Inpainting
Technique using diffusion models to coherently reconstruct missing or masked areas of an image based on the surrounding context.
Binary Inpainting Mask
Map of the same dimensions as the source image indicating which pixels to modify (value 1) and which to preserve (value 0), serving as a guide for the diffusion process.
Conditional Text Guidance
Method guiding the diffusion process for inpainting using a textual description, ensuring that the generated region respects the requested semantics.
Guided Resampling
Iterative noise addition and denoising strategy where noise samples are conditioned by the known pixels of the image to maintain consistency at mask boundaries.
Latent Blending
Fusion process in the latent space between the features of the original image and those generated by diffusion, ensuring a seamless transition.
Adaptive Timestep Diffusion
Approach where the number of denoising steps is dynamically adjusted based on the complexity of the region to inpaint, optimizing quality and computation time.
Context Propagation
Mechanism by which structural and textural information from the preserved areas of the image is propagated into the masked area during the diffusion process.
Structural Inpainting
Variant of diffusion inpainting focusing on reconstructing fundamental structures and contours before generating fine textural details.
Semantic Consistency Control
Set of techniques, often based on cross-attention networks, aimed at ensuring that the generated content is semantically logical relative to its environment.
Conditional Denoising by Mask
Step in the diffusion process where the U-Net model predicts the noise to be removed by jointly using the noisy image and the binary mask as conditional inputs.
Image Editing by Guided Diffusion
Application of diffusion inpainting not to fill a void, but to alter an existing region according to a directive (text, sketch, style) while preserving the rest.
Plug-and-Play Inpainting
Method allowing the use of a pre-trained diffusion model for inpainting without specific retraining, by modifying only its inference process.
Edge Replication
Technique for initializing noise in the masked area based on the characteristics of the border pixels, to improve the visual integration of the result.
Multi-scale Inpainting
Strategy performing diffusion at multiple resolutions, starting with a coarse structure to progressively refine details, improving overall consistency.
Anti-Border Blur
Post-processing or constraint applied during diffusion to avoid blur or misalignment artifacts at the junctions between the original area and the inpainted area.
Stochastic Diffusion Inpainting
Approach where the generation process introduces a controlled element of randomness, allowing for multiple plausible results for the same region to be completed.
Object Identity Preservation
Inpainting challenge involving modifying part of an object (e.g., a face) without altering its perceived identity, requiring fine control mechanisms.
Prompt-to-Region Inpainting
Advanced technique linking a specific text prompt segment to an image region, enabling localized and complex edits via diffusion.