Career NavigatorBY BIG BLOCK STUDIOS

AI TERMS FOR CREATIVE WORK

AI Glossary Speak

Plain-language AI terms and practical examples for creative work.

Your selection follows you across Career Resources.

LESS JARGON. MORE CLARITY.

What does that actually mean?

Search by term, abbreviation, or topic. Usage varies between teams, so ask how an employer uses a term.

Want to go deeper? Explore the reference glossaries ↓

Quality, rights & safety

Acceptance criteria

The specific conditions a result needs to meet before you use it.

In practice. Write down the required format, accuracy, accessibility checks, and brand rules before generating options.

Working method

Agentic workflow

A workflow where an AI system can take multiple steps or use tools toward a goal, within the permissions it is given. Autonomy varies.

In practice. Ask what actions it can take, where approval is required, and who handles errors.

Search & workflows

AI agent

A system that uses a model to choose and carry out steps toward a goal, often through connected tools.

In practice. Define which files it can access, what it can change, and when it should stop for review.

Working method

AI-assisted workflow

A process in which someone uses AI for particular tasks while retaining responsibility for how the output is used.

In practice. In a portfolio, identify the task, tool contribution, review, and your final decisions.

AI foundations

Artificial intelligence (AI)

Software that performs tasks such as recognizing patterns, making predictions, or generating content.

In practice. Ask which task the AI performs and how people check its output.

Quality, rights & safety

Benchmark

A shared test or dataset used to compare how systems perform on a task.

In practice. Pair published results with tests on your own briefs, assets, and quality standards.

Quality, rights & safety

Bias

A systematic tendency that can skew a model’s outputs or produce uneven results across groups or situations.

In practice. Review how generated imagery represents different people and settings across a set of prompts.

Origin & history

Content Credentials

A way to attach or associate provenance information with digital content. Supported tools and platforms differ.

In practice. Use alongside a clear case-study explanation. Credentials do not establish that all content is true or that every use is permitted.

Prompting & controls

Context window

The amount of information a model can consider at once, typically measured in tokens.

In practice. Put the most important instructions and references in reach for the task you are doing.

Team structure

Creative operations

The systems and coordination that help creative work move from request to delivery: intake, resourcing, approvals, and asset management.

In practice. Ask whether a role owns the process, creative decisions, production, or some combination.

Team structure

Creative pod

A small team organized around a client, project, or goal. Its disciplines and decision-making arrangements vary. It does not inherently involve AI.

In practice. Ask who belongs to the pod, who leads it, and whether it is permanent or project-based.

Quality, rights & safety

Data retention

How long a service keeps inputs, outputs, or logs, and under what conditions it deletes them.

In practice. Check the provider’s settings and your team’s agreement before entering client material.

AI foundations

Deep learning

A branch of machine learning that uses neural networks with multiple layers.

In practice. When a tool uses deep learning, ask what input it needs and what kinds of mistakes it tends to make.

AI foundations

Diffusion model

A generative model that learns to reverse a gradual addition of noise, often to create images.

In practice. Use it to explore visual directions and check the output at the final crop and size.

Search & workflows

Embedding

A list of numbers representing features of text, images, or other data so a system can compare them.

In practice. Use embeddings to find similar assets, then review the matches for brand or licensing fit.

Quality control

Evaluation / evals

A repeatable way to assess whether an AI system or output meets a set of requirements.

In practice. For creative work, define examples and criteria such as brand fit, technical quality, and factual accuracy.

Prompting & controls

Few-shot prompting

Showing a model a small number of examples of the behavior or format you want.

In practice. Provide two approved examples when you need a consistent headline structure or metadata format.

Models & training

Fine-tuning

Further training an existing model on a narrower set of examples for a particular task or behavior.

In practice. Try clearer instructions and reference examples first, then test whether fine-tuning would address the remaining gap.

AI foundations

Foundation model

A model trained on broad data and designed to be adapted to many different tasks.

In practice. Check which model powers your tool and what data-use terms apply to it.

AI foundations

Generative AI

AI that creates text, images, audio, video, code, or other content from inputs.

In practice. Treat generated concepts as drafts and check claims, rights, and brand fit before sharing them.

Search & workflows

Grounding

Tying an AI response to evidence or context, such as a trusted document or database.

In practice. Ask for the supporting passage and compare it with what the model says.

Quality, rights & safety

Guardrail

A rule, filter, or check intended to limit unwanted AI behavior.

In practice. Pair a rule against publishing confidential content with a human review before release.

Quality control

Human review

A person evaluates output against defined requirements before it moves forward. A review can be superficial unless responsibilities are clear.

In practice. Ask who reviews accuracy, brand fit, accessibility, and suitability—and who can stop delivery.

Search & workflows

Image-to-image generation

Creating a new image using an existing image as input, often with a text instruction.

In practice. Use an approved source image and review how closely the result retains the subject and layout.

Models & training

Inference

Running a trained model on new input to produce an answer, prediction, or generated asset.

In practice. When choosing a tool, measure how long it takes to return an output in your real workflow.

Search & workflows

Inpainting

Generating or replacing content within a selected region of an image.

In practice. Mask only the region you intend to change and inspect the seam at full size.

AI foundations

Large language model (LLM)

A model trained on large amounts of text to predict and generate sequences of language.

In practice. Use it to draft variations, then verify the facts and rewrite in your own voice.

Models & training

Latency

The time between sending a request and receiving a response.

In practice. Test response time on a typical brief before adding a tool to a live creative review.

Models & training

LoRA (low-rank adaptation)

A fine-tuning technique that trains a small set of added parameters while leaving most of the base model fixed.

In practice. Check that you have permission to use the training images and that the adapter works with your base model.

AI foundations

Machine learning (ML)

A way to build systems that learn patterns from examples and use them on new inputs.

In practice. Ask what examples the system learned from and where those examples differ from your audience.

Models & training

Model card

A document describing a model’s intended uses, evaluation results, and known limitations.

In practice. Scan the model card before using a new model for a client project.

Quality control

Model hallucination

An AI output that presents invented or unsupported information as though it were reliable.

In practice. Check names, quotes, claims, and references against original sources before delivery.

AI foundations

Multimodal model

A model that can work with more than one kind of input or output, such as text, images, or audio.

In practice. Check whether it can read your reference image and whether it can also create the output format you need.

Prompting & controls

Negative prompt

A field in some generation tools for describing elements you want the output to avoid.

In practice. Use it to discourage unwanted details, then inspect the output for anything that slipped through.

AI foundations

Neural network

A model built from layers of connected calculations that learn to transform inputs into outputs.

In practice. For image classification, check how well it handles unusual lighting, cropping, or unfamiliar subjects.

Models & training

Open-weight model

A model whose learned numerical weights are available to download under specified terms.

In practice. Read the license for commercial use, redistribution, and any restrictions.

Search & workflows

Outpainting

Extending an image beyond its original edges by generating new surrounding content.

In practice. Use it to adapt an image to a wider crop, then check perspective and continuity at the edges.

Models & training

Overfitting

When a model learns its training examples too closely and performs poorly on new examples.

In practice. Test a workflow with briefs and reference images that were never used to tune it.

Models & training

Parameters

The numerical values a model learns during training.

In practice. Compare actual results and running costs when choosing between models of different sizes.

Quality, rights & safety

Personally identifiable information (PII)

Information that can identify a person on its own or in combination with other information.

In practice. Remove personal details from briefs and examples before sending them to an external tool.

Prompting & controls

Prompt

The instructions and context you give an AI system for a particular task.

In practice. Include the audience, deliverable, constraints, and criteria you will use to judge the result.

Quality, rights & safety

Prompt injection

Untrusted instructions in a document, webpage, or other input that try to change how an AI system behaves.

In practice. Keep retrieved content separate from trusted instructions and require approval for sensitive actions.

Origin & history

Provenance

Information about where an asset came from and how it has been changed. A record may be incomplete.

In practice. Keep your own process notes and source records; do not treat missing metadata as proof of human authorship.

Models & training

Quantization

Using lower-precision numbers to make a model smaller or cheaper to run, sometimes at a cost to quality.

In practice. Compare small type, fine details, and consistency before switching to a smaller version.

Quality, rights & safety

Red teaming

Deliberately testing a system for harmful behavior, failures, or ways it can be misused.

In practice. Try out-of-scope requests before launching an internal assistant and document where it needs limits.

Search & workflows

Retrieval-augmented generation (RAG)

Retrieving relevant material at request time and giving it to a model to inform its answer.

In practice. Connect it to approved brand guidelines and check whether the answer points to the right passage.

Prompting & controls

Seed

A starting value for a generation tool’s random-number process.

In practice. Log the seed along with the model version and settings when an image iteration matters.

Search & workflows

Semantic search

Search that aims to match meaning even when the query and result use different words.

In practice. Try describing the asset you need, then check that the results actually fit the brief.

Prompting & controls

Structured output

A result constrained to a defined format, such as a JSON schema or a set of named fields.

In practice. Use it to collect a headline, summary, and alt text in fields your publishing workflow can check.

Content

Synthetic media

Images, audio, or video generated or materially altered using computational tools, including AI. Usage of the term varies.

In practice. Explain material generation or alteration where it affects how someone understands your work.

Prompting & controls

System instructions

Higher-priority instructions that shape how an assistant behaves across requests.

In practice. Use shared instructions for durable rules such as tone, confidentiality, and required checks.

Prompting & controls

Temperature

A generation setting that affects how broadly a model samples among possible next tokens.

In practice. Try a lower setting for repeatable formatting and test a wider range for brainstorming.

Prompting & controls

Token

A chunk of input or output a model processes, often a word fragment or punctuation mark.

In practice. When a long brief gets cut off, split it by task and carry forward the essential context.

Search & workflows

Tool calling

A way for a model to request that connected software look something up or perform an action.

In practice. Decide which tools can make changes and where a person needs to approve an action.

Models & training

Training data

The examples a model learns from while it is being built or adapted.

In practice. Ask about data provenance before using a model trained on client or third-party material.

AI foundations

Transformer

A neural-network architecture that uses attention to connect relevant parts of an input.

In practice. A transformer label describes how a model works. Test its output on the briefs you actually use.

Quality, rights & safety

Usage rights

The permissions and limits that govern how you may use, share, or license an asset.

In practice. Check the terms for both reference material and generated outputs before publication.

Search & workflows

Vector database

A system that stores numerical representations and supports searches for similar items.

In practice. If it indexes your asset library, check whose permissions apply to search results.

Quality, rights & safety

Watermark

A visible or embedded mark used to label content or signal where it came from.

In practice. Find out whether the mark survives cropping or export and whether disclosure is also required.

Prompting & controls

Zero-shot prompting

Asking a model to do a task without first showing examples of the desired output.

In practice. Start with a clear brief and add examples if the first result misses your format or tone.

Want to go deeper?

These larger glossaries explain the technical details and offer more examples.