AI Design Glossary

Technology that creates three-dimensional models from text or images using AI. It significantly accelerates the traditional 3D modeling process.

Technology that transforms static images or character designs into animated motion using artificial intelligence.

Artworks created using AI technologies or where AI plays an active role in the creative process. It enables producing digital art without needing traditional artistic skills.

Digital human representatives created with AI. They are used to produce video content with realistic facial expressions, speech, and gestures.

Technology that automatically removes or replaces backgrounds in photos using AI. It precisely separates the foreground through object segmentation.

Technology that automatically colorizes black-and-white photos and videos using AI. It predicts accurate colors through historical context and object recognition.

A broad field encompassing the integration of AI technologies into design processes. It includes applications ranging from UI/UX design to graphic design, architectural visualization to product design.

Technology that replaces a person's face in a photo or video with another face using AI.

The tendency of AI models to generate non-real, fabricated, or incorrect information/content. It manifests differently in text, image, and video generation.

A general term covering AI-powered photo and image editing tools. Includes automated retouching, object removal, style transformation, and more.

Inpainting edits/replaces specific areas of an image with AI, while outpainting extends image boundaries to create new content.

Technology that uses AI to redesign interior photos in different styles or create interior visuals from scratch.

Technology that automatically creates logos for brands and businesses using AI. It generates professional logo alternatives from text descriptions.

Technology that generates original music compositions, songs, and sound effects using AI. It creates music from text descriptions or melodies.

Technology that creates professional presentations using AI. It automatically generates slides, visual layouts, and design suggestions from text content.

Technology that creates professional product photographs or enhances existing product visuals using AI.

Advanced enhancement technology that uses AI to increase image resolution far beyond the original and reconstruct lost details.

Technology that uses AI to increase the resolution of low-resolution images. It intelligently predicts and reconstructs lost details.

Technologies that automate and accelerate the video editing process using AI. Covers automated cutting, color correction, subtitles, and effect application.

General term for technologies that create new video content from text, image, or video inputs using AI.

Technology that produces synthetic speech by cloning a person's voice using AI. It creates realistic voice copies from short audio samples.

The attention mechanism is an AI component that allows neural networks to selectively focus on different parts of input data.

The automatic processing of multiple images or operations simultaneously or sequentially. Instead of manually processing items one by one, it enables the management of hundreds or thousands of items through an automated pipeline.

A multimodal AI model developed by OpenAI that can represent text and images in the same vector space. Used as a prompt understanding layer in image generation tools.

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.

Cross-attention is a specialized attention mechanism where computations are performed between two different data sequences.

A technique that uses deep learning technology to realistically superimpose a person's face, voice, or movements onto another person.

Denoising is the fundamental principle of diffusion models; the model removes random noise step by step to form a coherent image.

A deep learning model that generates images by gradually denoising. It starts from random noise and step by step creates a meaningful image.

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 converting text, images, or other data types into dense, fixed-size numerical vectors. Used for semantic similarity calculation and model input representation.

Technology for improving and sharpening low-quality, blurry, or damaged face images using artificial intelligence. It automatically repairs facial components like eyes, mouth, and skin, producing naturally-looking results.

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

The FID Score is a standard evaluation metric used to measure the quality and realism of generated images.

A deep learning model where two neural networks are trained against each other: a generator and a discriminator. The generator tries to produce realistic data, while the discriminator tries to distinguish between real and fake data.

The general term for AI systems that can produce new and original content based on patterns learned from training data. It covers text, image, video, music, and code generation.

Guidance Scale controls how strictly generation adheres to the text prompt in diffusion models.

An AI technology that automatically describes the content of an image in text format. It expresses objects, scenes, colors, and relationships in the image with natural language sentences.

A technique that generates or transforms a new image using AI by referencing an existing image. The structure of the input image is preserved while style, content, or details can be modified.

Abbreviation for image-to-image. In the Stable Diffusion ecosystem, it refers to the mode of generating new images using a reference image. It transforms while preserving the structure of the original image.

The process where a trained AI model makes predictions or generates output on new inputs. In image generation, it corresponds to converting a prompt into an image.

A technique for regenerating or editing a specific area of an image by masking it with AI. Used for removing unwanted objects or modifying specific areas.

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

LCM reduces traditional diffusion models' dozens of steps to just 4-8 steps.

A multidimensional space where data is compressed and mathematically represented. Diffusion models perform image generation in this compressed space for computational efficiency.

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

A binary or grayscale layer in image processing that selects certain regions while excluding others. In operations like inpainting and segmentation, it defines which pixels will be changed.

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

A tool in video generation platforms that allows selecting specific regions of an image and defining custom movement direction and intensity for those regions. It enables creating more professional and intentional videos with selective motion control.

Multi-modal AI describes systems capable of processing different data types like text, images, audio, and video simultaneously.

A text command that defines unwanted elements in AI image generation. The model generates images while avoiding the elements specified in the negative prompt.

A technique that extends the boundaries of an existing image using AI. It enlarges the image by creating new areas that are consistent with the original content.

A text-based instruction or command given to AI models. It is used to describe the desired output and guides the model's generation process.

A discipline that encompasses techniques and strategies for writing prompts to get the best results from AI models. It includes proper word choice, structuring, and parameter usage.

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.

SDXL is an advanced diffusion model released by Stability AI in 2023, offering 1024x1024 native resolution.

A technique that applies the artistic style of one image while preserving the content of another. It has uses such as reinterpreting photographs in the style of famous painters.

Technology for creating a high-resolution version from a low-resolution image using artificial intelligence. Similar to upscale but can also fill in non-original areas of the image through hallucination.

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.

Technology that generates images from natural language text descriptions using artificial intelligence. The prompt written by the user is interpreted by the AI model and converted into an image.

Technology that generates video content from natural language text descriptions using artificial intelligence. It converts text prompts into moving, consistent frame sequences.

Technology that automatically generates video from text prompts using AI. It converts written descriptions into moving image sequences.

A technique for generating tileable textures or patterns where the edges of an image connect seamlessly with each other. Used in games, graphic design, and fabric design for creating repeating patterns.

The basic unit used by AI models when processing text. It can be a word, word fragment, or character. Prompt length and model capacity are measured in token count.

A deep learning architecture based on the attention mechanism with parallel processing capability. It forms the foundation of both language and visual models.

Abbreviation for text-to-image. In the Stable Diffusion ecosystem, it refers to the mode of generating images from text prompts. Unlike img2img, it produces images from scratch.

The process of enlarging low-resolution images using AI without quality loss or with quality enhancement. It differs from traditional resizing with its ability to add detail and sharpen.

A probabilistic deep learning model that encodes data into a compressed latent space and can generate new data from this space. Used in the image encoding layer of diffusion models.

The extension of diffusion models to the time dimension for use in video generation. By denoising in both spatial and temporal dimensions, it creates consistent and fluid video sequences.

Technology that transforms an existing video into a new one using AI while taking it as reference. The video's structure and motion are preserved while style, content, or visual quality is modified.

AI technology that detects visible or invisible watermarks in images and in some cases removes them. Used in content verification, copyright protection, and AI-generation tracking fields.

Zero-shot learning is the ability to perform tasks never seen in training data using only general knowledge.