A working glossary of the terms that recur in AI brand creative production — covering the AI tooling side, the print and OOH production side, and the workflow vocabulary that brand teams adopting generative tooling end up using.
AI Slop
The visual register of generative imagery produced by general-purpose models trained on broad image corpora. Recognizable markers include over-smooth textures, off-anatomy hands, drifting environmental detail, and the soft uncanny quality that distinguishes such output from professional brand creative.
Aspect Ratio
The proportional relationship between an image's width and height (e.g., 1:1, 16:9, 9:16). Brand creative typically requires the same parent composition to travel across multiple aspect ratios for different placements without losing identity.
Bleed
In print production, the area of an image extending past the trim edge — typically 3–5 millimeters — so that the cut does not leave a white edge. Print-ready files require bleed allowance built into the composition.
Brand Consistency
The property of brand creative output that holds visual register, talent direction, palette, and compositional logic across formats, channels, markets, and time. The hardest problem in generative creative production at scale.
CMYK
The four-color process (cyan, magenta, yellow, black) used in commercial print reproduction. Most AI image tools generate in sRGB, which has a wider gamut than CMYK and consequently loses information during conversion. Production-grade tools deliver CMYK output with the printer's specific ICC profile.
ControlNet
A neural network architecture used with diffusion models like Stable Diffusion to add precise control over generation — pose, composition, depth, edges, style transfer. The technique is widely used in production workflows that need predictable output across variants.
Diffusion Model
A family of generative AI architectures that produce images by progressively denoising random patterns into coherent outputs. The architectural basis for Stable Diffusion, Midjourney, DALL·E v3, and most production-grade AI image tools.
DOOH
Digital out-of-home — out-of-home advertising placements delivered through digital screens rather than traditional printed media. Includes roadside digital billboards, station screens, and transit displays.
Fine-tuning
The process of adapting a pre-trained model to a specific domain by continuing training on a smaller, more focused dataset. Per-brand fine-tuning on the brand's existing creative archive is the strongest available technique for maintaining brand visual register in generative output.
Format Suite
The full set of variant assets a campaign requires across placements — 1:1 social, 9:16 vertical, 16:9 horizontal, 6-sheet, 48-sheet, magazine spread, email banner, and so on. Production-grade workflows generate the suite from a single parent composition rather than producing each variant independently.
Generative AI
A category of AI systems that produce new content — images, text, audio, video — rather than classifying or analyzing existing content. Distinguished from discriminative models, which sort or score input.
ICC Profile
A standardized data set characterizing the color response of a specific device or process. Print production requires the file to be prepared in the printer's ICC profile so that screen-side color decisions reproduce correctly on press.
Image-to-Image
A generation mode that takes an input image and produces a transformed output rather than generating from text alone. Used in production for style transfer, environmental extension, and variant generation that preserves source composition.
Inpainting
A generation technique that modifies a specific region of an existing image while preserving the rest. Used in production for cleanup, asset replacement, and detail correction.
LoRA
Low-Rank Adaptation, a fine-tuning technique that adapts a pre-trained model with much smaller storage and compute requirements than full fine-tuning. Common in Stable Diffusion workflows for character, style, and brand-specific adaptation.
Megapixel
A unit of image resolution equal to one million pixels. Standard professional photography produces images at 12–24 megapixels. Production-grade AI tools deliver output at 12–16 megapixels to support print and large-format display.
OOH
Out-of-home advertising — placements outside the consumer's home, including billboards, transit, retail, environmental, and street-level displays. The category is among the most demanding for resolution, color, and production-prep.
Outpainting
A generation technique that extends an existing image beyond its original boundaries while preserving the central composition. Used in production for aspect-ratio adaptation and environmental extension.
Prompt
The text input that directs a generative model's output. The art and discipline of prompting — what to specify, what to leave open, how to combine references — is a distinct skill that has shaped how brand creative teams interact with generative tools.
Provenance
Documentation of the inputs, model versions, and steps that produced a given AI output. Required for regulator-facing audit, talent-rights documentation, and brand-side rights and compliance review.
Reference Image Conditioning
Using one or more reference images alongside a prompt to anchor a generation. The technique carries information that prompts alone cannot — composition, color, lighting, talent direction — and is central to production-grade brand consistency.
Rich Black
A printing convention where deep blacks are produced from a combination of CMYK inks (typically C: 60, M: 40, Y: 40, K: 100) rather than pure 100K black. Holds shadows against ink limit constraints and prevents mottling on press.
Seed
A numeric value that initializes the random component of a generative model. Locking a seed allows reproducible generation of a specific output, and is used in production to derive controlled variants from a parent asset.
Sora
OpenAI's generative video model, publicly unveiled in February 2024. The first widely-known generative video model capable of producing sixty-second clips with coherent identity, environment, and motion. Used in early brand experiments through 2024 and 2025.
sRGB
A standardized color space optimized for screen reproduction. The default output color space for most AI image tools. Must be converted to CMYK with appropriate gamut mapping for print production.
Stable Diffusion
An open-source family of diffusion models released by Stability AI starting in 2022. The architectural foundation for a wide range of brand-side and consumer-side AI image tooling, distinguished from closed alternatives by self-hosted deployment and customization flexibility.
Total Area Coverage (TAC)
In print production, the maximum combined percentage of CMYK ink coverage a coated stock can accept (typically 280–320%). Files exceeding the limit risk smearing or ink set-off on press. Production-prep handles TAC compliance.
Training Data
The dataset a model learns from. The character of generative output is a direct function of the training corpus. A model trained on the open internet produces output reflecting the open internet; a model trained on brand campaigns produces output reflecting brand campaigns.
Upscaling
Increasing the resolution of an existing image. AI-based upscaling can produce useful results for moderate increases but does not reliably invent detail at very large scale factors. Starting from higher native resolution is preferable to aggressive upscaling.