Local-first creative intelligence for mobile-game UA teams — decompose creatives into structure, link them into a graph, turn every human ruling into a compounding team asset.
-
Updated
Nov 11, 2026 - Python
Local-first creative intelligence for mobile-game UA teams — decompose creatives into structure, link them into a graph, turn every human ruling into a compounding team asset.
Open-source, self-hosted tool that scores any ad — image, video, or copy — across 8 brain regions using CLIP and Whisper, and names the weakest spot before you spend on it. CPU-only, FastAPI backend, Next.js frontend.
一条脚本铺到 10 个跨境渠道,不用每个市场重写一遍 | One script, ten marketplaces: cross-border content that clears the fact gate · 📮 wow.vip@foxmail.com · 🎬 https://v.douyin.com/07EjIDGVErY/
Count literal TikTok caption cues and draft human-reviewed experiment briefs; no video analysis or performance attribution.
Predicts where people look on an ad before it runs: classical + learned saliency, AOI focus scoring, and motion salience for video. Stress-tested on synthetic probes; static model scored on human eye-tracking (OSIE, 200 held-out images: CC 0.344 → 0.450, human ceiling 0.88). Natural scenes only, not yet validated on ads. Reproducible.
To associate your repository with the creative-testing topic, visit your repo's landing page and select "manage topics."