20 ms·
Stack/infra (high-level): - Frontend: React (Vite/Next) - Backend: Node/TypeScript + REST; queued renders - Inference: Nano Banana on Gemini Flash 2.5 Image on
by mesutgenai 1y ago
Stack/infra (high-level):
- Frontend: React (Vite/Next)
- Backend: Node/TypeScript + REST; queued renders
- Inference: Nano Banana on Gemini Flash 2.5 Image on Vertex;
- Infra: Docker/K8s; object storage for short-lived assets; CDN
Privacy/safety:
No training on user images; no resale/sharing of uploads
“AI Edit / Parody” watermark on exports; celebrity use is for parody/visualization only
NSFW/minors detection; blocklist for abusive inputs;
Limitations:
Best with front-lit, uncovered faces; large occlusions/sunglasses degrade results
Extreme poses and complex hair can produce edge artifacts
Roadmap:
Side-by-side before/after, bridal templates, batch renders, higher-res, pose-match hints
Happy to share more details or logs/screens if useful.