Photography AI Glossary

Plain-English definitions for AI photo management terms.

LrGeniusAI brings concepts from AI and machine learning into your Lightroom workflow. This glossary explains the key terms you'll encounter — no prior AI knowledge required.

AI Keywording

The automated generation of descriptive tags (keywords) for photos using an AI vision model. Unlike manual tagging, AI keywording analyzes the visual content of each image — subjects, scenes, colors, actions — and writes keyword metadata directly to the photo. LrGeniusAI uses open-vocabulary keywording, meaning it generates any relevant term rather than picking from a fixed list.

See also: Analyze & Index guide

Semantic Search

A way of searching a photo library by meaning or description rather than exact keyword matches. You describe what you remember — 'child running on a beach at sunset' — and the AI finds visually matching photos, even if those exact words were never used as keywords. Semantic search works by comparing the meaning of your query against the AI's understanding of each indexed image.

Culling

The process of selecting the best photos from a shoot and rejecting the rest — blurry shots, duplicates, eyes-closed frames, unflattering expressions. AI culling automates this first-pass selection by analyzing each image for sharpness, face quality, and composition. In LrGeniusAI, culling is face-aware, meaning the AI factors in whether eyes are open and in focus.

Embedding

A numerical representation of an image (or a piece of text) that captures its meaning as a list of numbers called a vector. AI search and semantic similarity work by comparing embeddings: if two photos have similar embeddings, they are visually or semantically similar. When LrGeniusAI indexes your library, it generates and stores an embedding for each photo so that semantic searches can run instantly later.

See also: Analyze & Index guide

Vision-Language Model (VLM)

An AI model that understands both images and text simultaneously. Instead of only recognizing objects from a fixed list, a VLM can describe what it sees in natural language, answer questions about an image, and generate contextually rich keywords and captions. LrGeniusAI supports multiple VLMs — models it runs itself (Gemma 4, Qwen3-VL, Qwen2.5-VL), models served by Ollama or LM Studio, and cloud models like GPT-4o and Gemini.

llama.cpp

An open-source C++ inference engine for running language and vision models efficiently on ordinary consumer hardware. LrGeniusAI embeds llama.cpp directly in its backend, so it can run a local vision model in-process without Ollama, LM Studio, or any other AI app installed. In LrGeniusAI it is the engine that ships on Windows, with GPU acceleration via Vulkan; Apple silicon Macs use MLX instead.

MLX

Apple's machine-learning framework built specifically for Apple silicon, using the unified memory architecture of M-series chips to run models efficiently on the GPU. On Apple silicon Macs it is the built-in engine LrGeniusAI ships, running vision models through a small Metal helper process; Windows uses llama.cpp instead.

GGUF

The model file format used by llama.cpp — a single file containing the model weights, quantized to reduce size and memory use. For vision models a GGUF comes as a pair: the weights plus an 'mmproj' vision projector, the part that lets the model actually see an image. LrGeniusAI downloads both automatically when you pick a local model.

Quantization

Compressing an AI model by storing its weights at lower numerical precision — for example 4 bits per weight instead of 16. A quantized model uses far less RAM and runs faster, at a small cost in output quality. Quantization is what makes it practical to run a capable vision model on a laptop, and it is why LrGeniusAI's local models are a few gigabytes rather than tens of gigabytes.

Ollama

An open-source runtime for running large language and vision models on your own hardware. Ollama handles model downloading, management, and a local API server, so applications like LrGeniusAI can send images to a local AI model without any internet connection. LrGeniusAI can also run local models itself via llama.cpp and MLX, so Ollama is an option rather than a requirement — useful if you already run it for other tools.

LM Studio

A desktop application for running AI language and vision models locally on your own hardware, with a user-friendly interface for model management. LM Studio provides a local API compatible with the OpenAI format, which LrGeniusAI can connect to as an alternative to Ollama. LrGeniusAI also reuses models already downloaded in LM Studio when running them through its own built-in engines, so you are not storing a second copy.

Open Vocabulary Keywording

AI keywording that generates any descriptive term the model considers relevant, rather than selecting from a predefined, closed list. Most traditional photo keywording tools use a fixed taxonomy — you can only get keywords that were explicitly included. Open-vocabulary models describe images in natural language, producing richer, more specific, and more accurate keywords for unusual subjects or niche genres.

See also: Analyze & Index guide

CLIP

Contrastive Language-Image Pretraining — a foundational AI model architecture developed by OpenAI that learns to understand images and text in a shared embedding space. Models trained with CLIP-style objectives can match images to textual descriptions and vice versa, which is the basis for semantic image search. Many vision-language models in LrGeniusAI build on CLIP-style training.

Face Recognition

The automatic detection and identification of people across a photo library. LrGeniusAI's face recognition groups photos by the people who appear in them, building a People cluster so you can quickly find all photos of a specific person without manual tagging. In culling mode, the AI also uses face detection to prioritize frames where the subject's eyes are sharp and open.

AI Caption

A natural-language sentence or short paragraph describing what is in a photo, generated by a vision-language model. Captions are more descriptive than keywords and can convey narrative context — 'A photographer adjusts camera settings on a tripod at dawn with mountains in the background.' LrGeniusAI can generate captions alongside keywords and write them to Lightroom's caption metadata field. Captions are also used as alt-text for web accessibility.

See also: Analyze & Index guide

Lightroom Classic Catalog

Adobe Lightroom Classic's database file that tracks the location of your photos, stores all edits, metadata, keywords, ratings, and organizational structure. Unlike Lightroom (cloud-based), Lightroom Classic keeps the catalog locally on your computer. LrGeniusAI is a plugin for Lightroom Classic that reads and writes keyword metadata and other fields directly into this catalog.

Keyword Deduplication

The process of identifying and merging duplicate, redundant, or near-identical keywords in a Lightroom catalog. Over time, catalogs accumulate variants like 'dog', 'dogs', and 'Dog' as separate entries — deduplication consolidates these into a single canonical keyword and re-applies it across all affected photos. LrGeniusAI includes a Keyword Dedup & De-Clutter tool for this purpose.