AI & Machine Learning
Glossary

A broad field of computer science focused on building systems capable of performing tasks that typically require human intelligence, such as visual perception, speech recognition, decision-making, and language translation.

A technique that allows models to focus on relevant parts of the input when generating each part of the output. Enables models to capture long-range dependencies in data.

Generating multiple outputs in a single request or processing multiple inputs together. More efficient than making individual requests for each generation.

Systematic errors in AI outputs due to biases in training data or model design. Can result in unfair or inaccurate representations of certain groups or concepts.

A parameter that controls how closely the generated image follows the prompt. Higher values produce outputs more aligned with the prompt but may reduce diversity and quality.

A saved state of a model's weights at a point during or after training. Allows resuming training, comparing versions, or using the model at different stages of development.

A model trained by OpenAI that learns to associate images with text descriptions. Used in image generation to guide the creation process based on text prompts.

Systems and practices to prevent AI from generating harmful, illegal, or inappropriate content. Includes filters, classifiers, and model training restrictions.

A neural network architecture that adds conditional control to diffusion models. Allows guiding image generation with additional inputs like edge maps, depth maps, poses, or sketches.

OpenAI's image generation model series. DALL-E 3 is known for excellent prompt understanding and safe, high-quality outputs. Available through OpenAI's API.

A subset of machine learning that uses neural networks with many layers (hence 'deep') to learn complex patterns in data. Powers most modern AI image, video, and audio generation.

A type of generative model that learns to create images by reversing a gradual noising process. Starting from pure noise, the model iteratively removes noise to generate coherent images. Used by FLUX, Stable Diffusion, and DALL-E 3.

A fine-tuning technique that teaches a model to generate specific subjects (people, objects, styles) from just a few example images. Associates a unique identifier token with the subject.

Adapting a pre-trained model to a specific task or domain by training it further on a smaller, targeted dataset. More efficient than training from scratch.

A family of high-quality image generation models by Black Forest Labs. Known for excellent prompt following, text rendering, and diverse outputs. Available in Schnell (fast), Dev (balanced), and Pro (quality) variants.

Generating intermediate frames between existing video frames to increase frame rate or create smooth slow-motion effects. AI predicts the motion between frames.

A generative model architecture where two neural networks (generator and discriminator) compete against each other. The generator creates fake samples while the discriminator tries to distinguish them from real ones.

AI systems that can create new content, including images, text, audio, and video. These models learn patterns from training data and generate novel outputs based on prompts or inputs.

AI generation that takes an existing image as input along with a prompt to create a modified or transformed version. Used for style transfer, editing, and variations.

Converting a static image into a video by generating motion and animation. The AI predicts plausible movement based on the image content.

The process of using a trained model to generate outputs from new inputs. When you call an AI API to generate an image, you're running inference on the model.

The process of filling in or replacing specific regions of an image while maintaining coherence with the surrounding content. Used for removing objects, editing specific areas, or extending images.

The time between sending a request and receiving a response. For AI generation, latency includes queue time, inference time, and network transfer.

A compressed representation of data learned by AI models. In image generation, manipulating points in latent space allows for smooth transitions between generated images and control over image attributes.

A technique for efficiently fine-tuning large models by training small adapter layers rather than all model weights. Produces small files that can be combined with base models.

A subset of AI that enables systems to learn and improve from experience without being explicitly programmed. ML algorithms build models based on training data to make predictions or decisions.

A popular AI image generation service known for its artistic, stylized outputs. Accessed through Discord. Not available through APIs.

AI systems that create original music compositions from prompts, existing audio, or musical notation. Can generate melodies, harmonies, and full arrangements.

Text describing what you don't want in the generated output. Helps avoid common artifacts or unwanted elements. Supported by most diffusion models.

A computing system inspired by biological neural networks. Consists of interconnected nodes (neurons) organized in layers that process information and learn patterns from data.

Extending an image beyond its original boundaries by generating new content that seamlessly continues the existing image. Also called image extension or uncropping.

The text input given to a generative AI model to guide what it creates. Effective prompting is key to getting desired results from AI generation.

The practice of crafting effective prompts to achieve desired outputs from AI models. Involves understanding model behavior, using specific keywords, and iterative refinement.

The algorithm used to generate images in diffusion models. Different samplers (Euler, DPM++, DDIM) offer trade-offs between speed, quality, and style.

A number that initializes the random number generator, allowing reproducible generations. Using the same seed, prompt, and settings produces identical outputs.

AI systems that transcribe spoken audio into written text. Also called automatic speech recognition (ASR). Used for transcription, subtitles, and voice interfaces.

An open-source latent diffusion model for image generation developed by Stability AI. The most widely used open model, with many community fine-tunes and extensions.

In diffusion models, the number of denoising iterations used to generate an image. More steps generally produce higher quality but take longer. Typical range: 20-50 steps.

The coherence of generated video across time—ensuring objects, lighting, and style remain consistent from frame to frame without flickering or sudden changes.

AI systems that generate images from text descriptions (prompts). Models like FLUX, DALL-E, and Midjourney interpret natural language to create corresponding visual content.

AI systems that convert written text into natural-sounding speech. Modern TTS can clone voices, control emotion, and produce highly realistic audio.

AI systems that generate video clips from text descriptions. These models create coherent motion and temporal consistency across frames while interpreting prompts.

The basic unit of text that AI models process. Words, parts of words, or punctuation marks are converted to tokens. Models have limits on the number of tokens they can process.

The process of teaching a machine learning model by showing it examples and adjusting its parameters to minimize errors. Requires large datasets and significant computational resources.

A neural network architecture that uses self-attention mechanisms to process sequences. Powers both large language models and modern image/video generators. Key innovation: parallel processing of entire sequences.

A neural network architecture originally designed for image segmentation, widely used in diffusion models for the denoising process. Features skip connections between encoder and decoder layers.

Increasing the resolution of an image using AI to add detail and clarity. AI upscalers can generate plausible details that weren't in the original low-resolution image.

A neural network architecture that learns to compress data into a latent space and reconstruct it. VAEs are used in diffusion models to work in compressed latent space rather than pixel space, improving efficiency.

Creating a synthetic voice that mimics a specific person's voice characteristics. Requires training on audio samples of the target voice.

Embedding invisible or visible identifiers in AI-generated content to indicate its synthetic origin. Helps with authenticity verification and content provenance.