01The problem
A static creative passes through different hands and tools: whoever writes the concept, whoever turns it into an image, whoever uploads the ad by copying texts, formats and names. Every step is a chance to lose time or introduce a mistake and, multiplied across several clients, it becomes the bottleneck of testing.
02How it is used
- You start from the brief: client, offer, goal and the page or texts to work from.
- You choose among three concepts. Each one draws on our libraries of winning ads and on the analysis of the ads competitors have running: headline, subheadline, call to action and visual directions, within the client's brand system.
- Before the image, the layout. The composition is approved on a sketch: correcting a layout costs nothing, regenerating an image does.
- You approve the image. Only the approved one goes on, identified by a unique fingerprint of the file.
- The ad is born paused. It is uploaded to Meta only after the OK on a summary and stays still until a person turns it on.
03How it is built
- Conceptsa language model writes the proposals starting from the client's brand system, our libraries of winning ads and the analysis of competitors' ads; a check on the text discards generic formulas
- Imagesgenerated through image-model APIs, only on explicit request
- Approvalevery image is tied to its fingerprint: exactly the approved file is uploaded, not a later version
- Uploada manifest lists campaign, ad set, texts and file; after the OK the script creates the ads via the Meta API, always paused
- Complianceon realistic images with people the “Image generated with AI” label is applied, as the AI Act requires
04Results
The first complete round reached Meta in October 2026: concept, image and paused ad, without copying texts or names by hand. Turning it on remains a manual decision, by design.