How AI‑Powered Creative Stitching Supercharges Ad Production

Introduction

Intelligent Creative starts with advertising assets, aiming to improve creative quality, scale, delivery effectiveness, and production efficiency through a programmatic AI system that serves advertisers, businesses, and platforms.

Business Status

Background

Various traffic sources require a large number of diverse creatives; creative quality and quantity directly affect traffic growth and cost control.

Business Pain Points

Advertisers lack the ability to produce high‑quality creatives.

Creative scale is limited despite diverse user scenarios.

Optimization relies on manual design and experience, lacking data‑driven guidance.

The entire workflow depends on design and operations manpower without systematic support.

Business Idea

Adopting a “one‑to‑many” design approach enables more precise ad delivery, enhancing user experience and business efficiency.

Core Capabilities

The tech team added AI‑driven capabilities such as intelligent stitching, dynamic matching, and creative selection, focusing here on intelligent stitching.

Preliminary Validation

Early experiments on local directory listing images showed positive data, confirming the direction.

Solution

How to Stitch Better

Intelligent stitching involves two key steps: decomposition (splitting elements) and programmatic recombination (stitching).

Element Decomposition (Split)

Using a DPA template as an example, elements include background, background decoration, car, car series name, price, call‑to‑action button, and button copy.

Content Decomposition (Split)

Content such as primary copy, secondary copy, car‑series copy library, price library, and hot‑tag library are also broken down.

Stitching Sizes

Analysis of ad sizes and positions yields four common image aspect ratios.

Stitching Styles

Single‑image illustration style.

Multi‑image real‑scene stitching style.

Stitching Layouts

Different styles emphasize different layout aspects: multi‑image real‑scene focuses on image count, ratios, and combination rules; single‑image illustration focuses on element placement (left/right split, diagonal, vertical split, symmetry, etc.).

Style Combinations

Combination 1 targets single‑image illustration, mixing layout, texture, color, and shape dimensions. Combination 2 targets multi‑image real‑scene, adding masks, layout choices, image count, overall shape, and auxiliary element styling.

Stitching Results

Cross‑dimensional templates generate numerous ad assets for business use.

Technical Integration

Source files must have standardized layer annotations; proper hierarchical relationships between layers are essential.

Stitching Applications

Current applications include local directory listings, DSP delivery, DPA delivery, and ongoing structuring of alliance ads into templates for systematic, intelligent output.

Verification

Production Capacity

In a real‑scene stitching scenario, two people working three days manually produce up to 200 images, whereas the AI stitching system can generate over 150,000 images, with the gap widening as more variation dimensions are added.

Data Performance

CTR improved across categories for directory listing images; DSP and DPA results await further data collection.

Conclusion

The intelligent creative system is rapidly maturing, scaling its capabilities and boosting efficiency. Future work will refine stitching logic based on size ratios and visual interfaces, further freeing designers to focus on higher‑value work and delivering greater business value.