WebinsaneAI production
Generation is cheap. Judgement isn’t.
AI made making nearly free. It moved the cost to deciding, and that is where the system has to be built.
Last month we rewrote the AI section of our own site. The first draft was a capability matrix of image, video, code and brand, and on rereading it said nothing a buyer would care about. The version that shipped makes one promise instead: on brand, at volume. Full campaign sets in a fraction of the usual time, each asset signed off by a person before it goes out.
A promise like that is really a claim about a process. So this is the process behind our AI content production, and the thinking that shaped it.
The cost moved; it didn’t disappear.
With current image and video models, a candidate asset takes seconds to produce. A brand, meanwhile, needs exactly what it always needed: the right idea, a consistent look, the correct product, cleared rights, the right size for every channel. AI has not removed that work. It has moved the cost of it, from making to directing and choosing.
Most of the disappointment I hear about AI production comes from teams who built for the old cost. They set up generation and expected the rest to follow. A production line designed for AI starts from the other end: it assumes generation is nearly free and invests everything in the decisions around it.
Consistency is decided before the first generation.
Every run on our line draws on the same brand pack. Approved reference images for product, people and places. Character sheets for anyone who recurs, so a face holds from asset to asset. The reasons are in The face is the brief. Palette, grade and composition written as constraints rather than moods. And voice rules for any copy, in every language the campaign runs in.
In practice, that pack is the brand’s design system, rewritten for a new kind of reader. The more precisely it is written, the less correction happens downstream.
Approve the picture before multiplying it.
We generate and art-direct a small set of hero stills and get them signed off before anything is animated or multiplied. A still is cheap to change. Two hundred variants of the wrong idea are not. This one sequencing decision saves more time than any model upgrade we have made.
Only once the key frames are approved does the line produce volume: the full set of formats from vertical to widescreen, product variants, language versions, channel cutdowns, and short video shots animated from the approved stills. This is the stage where days replace weeks, and the only stage most people picture when they hear the phrase.
A named person signs off every asset.
Every asset passes a person before it ships, and that person has a name. They check identity, product accuracy, hands and text, brand fit, and anything the model invented on its own initiative. We design the review like a product: assets grouped by type and shown beside their reference, so one reviewer can approve two hundred in an afternoon without approving a single bad one.
Rights and boundaries are settled at the start, not discovered at the end: whose likeness may be generated and with what consent, which models are cleared for commercial use, what each market requires to be disclosed, and what is simply off-limits for the brand. These are part of the brief the line runs on. Human sign-off is not a disclaimer at the bottom of the page. It is a requirement of the design.
Code is on the line too.
The same approach ships the sites, landing pages and tools a campaign runs on. AI coding agents work inside a documented system, and a developer reviews their work the way an art director reviews assets. This website was made that way: WebGL, scroll-driven, and built in weeks rather than months.
What it is not.
It is not prompt in, campaign out. It does not replace the idea, which still comes from people who understand a brand and its audience. And it has limits we state up front: long-form film with dialogue and continuity is still better shot. What it is, is the most reliable way we know to give a brand the volume modern channels demand without the work drifting off-brand. If that is the month you are facing, tell us about it.
Generation is cheap now. Judgement isn’t, and it shouldn’t be.
Asked often.
What is AI content production?
Producing brand images, video, copy and code with generative AI inside a directed workflow: brand inputs, approved key frames, multiplication into formats and languages, and human sign-off on every asset.
How do you keep AI-generated content on-brand?
By constraining the inputs rather than describing the outputs. That means approved references, character sheets, palette and composition rules and voice guidelines, and approving key stills before producing variants.
How fast can an AI production team deliver brand assets?
Once key frames are approved, full format and language sets of a few hundred assets can be delivered in days rather than weeks, including human review.