AI ad creative editor

The situation

Advertising creative is repetitive work: the same product, rendered again and again across dozens of placements and sizes. Marketing teams either send every variation back to a designer or settle for generic templates. The brief was an editor a marketer could drive themselves, with the tedious image work handled by models rather than by hand.

What we built

• An editor shell with a full tool system — select, draw, shapes and text — plus undo and redo, snapping and zoom
• A layers panel and a context-sensitive properties panel that changes with the selection
• A text tool with real typography controls: family, size, weight, alignment, letter-spacing and line-height
• Mask brushing and overlay tools for compositing
• An AI editing pipeline — background removal, layer extraction from flat artwork, and inpainting — returning results as ordinary editable layers
• A batch grid for generating many variants at once, with A/B comparison between them
• Filters, effects and an export dialog covering every required ad size

Engineering notes

The difficult part was putting the AI pipeline inside a normal editor rather than beside it. A background removal cannot return a flattened image — it has to come back as an editable layer carrying its own mask, so the user can keep working on it like anything else they drew. Several background-removal and segmentation models were compared on real product photography before one was chosen, because the quality difference on fine edges — hair, glassware, packaging — was large enough to decide the architecture.