What problem does it solve? Launching A/B tests on Meta, TikTok, Google, or Pinterest requires dozens of on-brand ad variants across multiple aspect ratios, and producing them one by one is slow and inconsistent. This Skill fans out a single approved hero image into a structured variant matrix with deterministic naming ready for direct ad-platform upload. ## Core Features & Use Cases - Variant matrix generation: Define 3-4 explicit axes (visual direction, composition, color, setting) and emit one batch job per variant with the hero as a reference image on every job. - Parallel batch execution: Run 15-30 image jobs at concurrency 6-8 via gen-ai batch run, with dry-run cost estimation and resume support for failed jobs. - Model routing per sub-task: Choose recraftv4 for brand consistency, flux-2-pro for photoreal ads, ideogram-v3 for baked-in headline text, or gemini-3.1-flash-image for cheap draft exploration. - Use Case: A marketer has one approved product hero and needs 15 variants (5 visual directions x 3 aspect ratios) for a Meta ad-set launch, named {campaign}_{visual}_{composition}_{size} for direct CSV import into Ads Manager. ## Quick Start Fan out 15 ad variants from my hero image hero.webp across 9:16, 1:1, and 16:9 for a Meta campaign launch.
