The premise: our data is stories
Most AI image platforms are trained on the open internet. That is fine if you want to generate a stock photo of a person holding a coffee cup. It is a problem if you are trying to produce brand creative, because the open internet is mostly not brand creative. It is photography, screenshots, fan art, memes, watermarks, and a great deal of work the original creators never intended a model to consume.
CampaignsLive starts from a different premise. The system learns from real brand campaigns — work that real teams briefed, paid for, and launched. The dataset is a record of what successful commercial creative actually looks like, not a guess at what creative might look like.
What we trained on
More than five hundred thousand brand campaigns launched globally since 2019, spanning:
- Every major category — automotive, financial services, retail, FMCG, telecom, hospitality, B2B, public sector.
- Every major media market — the dataset is not Western-centric. East Asia, Latin America, MENA, and SEA campaigns are present in proportion.
- Every standard channel — paid social, display, programmatic, OOH, print, broadcast, in-store, environmental.
- Five years of visual language — campaigns shift with culture and platform. A 2019 social tile reads differently from a 2024 one, and the model knows the difference.
The training data does not include the open internet at large, stock libraries, or user-generated content. It is a curated, brand-creative-only corpus.
How generation works
From the user's side, the workflow is intentionally close to how a creative team already operates:
- Brief. Describe the campaign in the language you would use internally — audience, message, brand, channel, reference points. Natural-language briefs work; structured briefs work; a tagline plus a moodboard works.
- First pass. The system generates a set of directions. These are starting points, not finished comps. The variation between them is in composition and approach, not in random style.
- Refine. Iterate on the directions that resonate. You can change talent direction, palette, environment, mood, or composition independently. The parent idea holds across iterations.
- Variants. Once a hero direction is approved, generate the format suite — every aspect ratio, every placement, every market — from the same parent. Each variant respects the parent rather than reinterpreting it.
- Export. Take the production-ready files in the format and color profile your downstream workflow expects. For print and OOH, that means CMYK with the standard profile your printer uses. For digital, that means sRGB at 16 megapixels.
What you get, in production terms
Every image is delivered at 16-megapixel resolution. That is enough for:
- Every digital placement at native resolution.
- Magazine spreads at 300 PPI without upscaling.
- Standard out-of-home formats from 6-sheets through 96-sheets.
- Environmental and wallscape placements with optional upscaling.
Video output is in active development. The roadmap target is 6K resolution for image-to-video, concept-to-clip, and full creative-to-spot workflows.
Color is delivered in the profile your placement requires — sRGB for screen, CMYK with your printer's profile for print. See Print and Out-of-Home for the full production-prep workflow.
IP ownership, in plain language
Everything CampaignsLive produces for your account is yours.
That sentence is short on purpose. The longer version: you own the creative output, the variations you iterate from it, and the format suite you generate from it. There is no separate licensing layer. There is no commercial-use restriction. There is no required attribution. There is no time-bound usage window. The file you export is treated, contractually, the way a file from a freelance art director or a contracted production house would be treated: yours, for any commercial purpose, indefinitely.
This is unusual in the current AI creative tooling market. Most platforms either retain rights to generated output, restrict commercial use, or license output back to the user under terms that survive cancellation. The motivation for taking a different position here is straightforward: brand teams cannot build long-term campaigns on assets they do not control. If the asset can be revoked, modified, or re-licensed by the platform, it is not a brand asset.
What CampaignsLive is not
It is worth being specific about scope:
- Not a stock library. The output is generated for your brief; it is not retrieved from a catalogue and is not shared with other users.
- Not a general-purpose image model. The training data is brand creative. The output domain is brand creative. Asking it for a meme template or a stock-style portrait will work less well than asking it for a campaign hero.
- Not a replacement for a creative team. The system shortens the path from brief to file. It does not write the brief, choose the audience, or sign off on the campaign. Those remain where they should remain.
Where this fits in a real production pipeline
Most teams adopt CampaignsLive in one of three places in the pipeline:
- Pre-production exploration. Visualizing a brief in a day instead of a week, to align stakeholders before committing to a shoot.
- Always-on creative refresh. Generating weekly performance variants without queueing studio time for work that does not warrant a shoot.
- Full campaign production. Replacing the shoot for campaigns where the creative direction is achievable generatively — a growing share of work as the output quality bar rises.
The earliest place to start is usually the first one — exploration. Teams that use the system for visualization in pre-pro tend to expand into production work organically as confidence in the output builds.
Getting started
The platform is open for new accounts. Use cases broken out by channel are covered in the Digital Advertising and Print and Out-of-Home pages. The blog publishes deeper notes from the training data, the workflow, and what the archive says about creative that performs.