AIARTDIRECTION.COM / THE VOCABULARY
The vocabulary of AI art direction, defined. Every term is written to be quoted: short, precise, and honest about what the discipline actually does. 21 terms, growing.
01 · THE DISCIPLINE
The practice of directing generative AI models across image, video, and sound to a professional creative standard. The art director sets visual intent, translates it into model instructions, enforces consistency, and curates and finishes the output. The full definition.
§A creative professional who directs generative AI models the way a conventional art director directs a photographer, stylist, and retoucher. They own visual intent, consistency, curation, and finish. How to become one.
§The senior rank above AI art director. Owns the idea and strategy across campaigns and teams, where the art director owns the execution of a project. Mirrors the traditional agency ladder.
§Writing and refining instructions to get better output from an AI model. One tool inside AI art direction, not the discipline itself. Prompt engineering optimises the instruction. Art direction owns the outcome.
§02 · THE CRAFT
The pre-generation phase where a project's visual rules are set: lighting, palette, materials, grade, and casting. What a frame is allowed to look like, decided before any frame exists. Borrowed from film and animation pipelines.
§Keeping face, product, wardrobe, and lighting continuous across every frame of a campaign. The difference between a lucky image and a deliverable, and the thing brands actually pay for.
§Holding the same recognisable person across many generated frames: face, build, hair, energy. The most requested skill in commercial AI work and the first thing a brand checks in a portfolio.
§The ability to read an image the way a director does: name its light source, lens behaviour, materials, palette, and era precisely enough to rebuild or brief it. The core skill taste is built on.
§Light in a frame that comes from an identifiable source: a window, a streetlamp, a flash. Undirected AI output defaults to sourceless even light. Directed frames name the source and let it cast real shadows. The Slop Test trains you to see the difference.
§The single frame that carries a campaign or shot. It gets art directed hardest, finished deepest, and is often the anchor input for video generation.
§A self-initiated campaign built inside a real brand's existing visual world without a commission. The standard portfolio proof that an AI art director can serve a brand, not just make pretty images.
§The post-generation pass that closes the gap between an AI render and a deliverable: colour grading, grain, retouch, cleanup, and upscaling. No professional frame ships straight out of the model.
§Adjusting colour, contrast, and tone across a set of frames so they read as one world. In AI art direction, grading doubles as a consistency tool: it unifies frames the model produced separately.
§03 · THE OUTPUT
Low-quality AI-generated content produced in volume without human direction. Merriam-Webster's 2025 Word of the Year, and the American Dialect Society's in January 2026. In commercial visual work the tells are default lighting, plastic surfaces, invented background text, and no single idea. Slop is not a model problem. It is the absence of an art director. Can you spot it?
§Frames generated versus frames shipped. Professionals generate wide, reject hard, and ship few. A high ratio is a sign of standards, not waste.
§04 · THE TOOLKIT
Generating a still image from a written description. The base capability of image models such as Midjourney, Flux, and Nano Banana.
§Generating a moving shot from a still frame plus a motion instruction. The standard professional video workflow in 2026: direct the still hard, then animate it.
§Regenerating a selected area inside an existing image while leaving the rest untouched. Used to fix details, swap products, or clean artefacts without regenerating the whole frame.
§Extending an image beyond its original borders, with the model inventing coherent surroundings. Used to change aspect ratios or widen a composition.
§Increasing an image or video's resolution with a model that reconstructs detail. The finishing step that takes a frame from screen preview to print grade. Creative upscalers can invent texture, which needs directing like everything else.
§A small trained add-on that teaches a model a specific face, product, or style. Short for low-rank adaptation. The heavy-duty answer to consistency when reference images alone will not hold.
§NO TERMS MATCH. THE DISCIPLINE IS YOUNG: SUGGEST A TERM VIA THE CENSUS LIST.
A term missing? The glossary grows with the discipline. Join the list and reply to any email with the vocabulary you hear in the wild.
Join the list