Social media ad generator

The Concepts ruler runs from 1 to 100 and starts at 8. Each concept is a separate billed render, so cost tracks the number you set. A batch is not one ad repeated: every concept gets its own ad style, hook, tone and scene, and each headline is claimed against the batch, so a concept landing on a headline a sibling already took gets rewritten.

Yes, once a project holds one. The References panel carries three slots, Product, Person and Backdrop, and it wants a real project, since Quick tools has nowhere to keep them. The render prompt then tells the model to match that exact container form, lid, label artwork and colors. Ad styles that are not product shots push the photo out of the first slot rather than drop it, so the label still comes back right.

An image at the aspect ratio you picked, with the headline and call to action typeset into it. Headlines are capped at 90 characters and calls to action at 32, so the copy fits the frame instead of getting clipped. You also get body copy for the post caption, which is not baked into the image, plus the ad style, hook type and pillar that concept was written for.

The Model row is named after the model that actually runs, gpt-image-2 for image ads, and marked Default for this tool. Leave it there and the tool keeps choosing; pick another image model from the list to pin one. The model only decides pixels. Headline, body copy and call to action are written first, then each concept is rendered with your references attached.

No. Every control is live while you are signed out, and Generate is where you are asked to sign in. Point the Project selector at a real project and the render reads that project's brand, research, strategy and reference pack, which is what makes the copy specific. Pick Quick tools and it still runs, but references have nowhere to live, so it works from your brief alone.