PoLang's smart search and AI chat rely on on-device models and a one-time gallery scan. Finish the steps below so the app can actually understand you.
Model downloads, account, people, and remote channels live in the app's Settings; the scan entry is at the bottom tab "Gallery Cleanup → Scan". Each step below points you to the right screen — the UI may differ slightly in the latest version.
Tagging, face recognition, and semantic search all rely on on-device AI models. Go to Model Center → Required and download them all (7 models, ~1.5 GB). The Recommended models (matting, voice) are optional.
Let the app learn the people and scenes in your gallery. Go to the bottom tab Gallery Cleanup → Scan and start scanning (face detection + semantic encoding, ~1.5h / 10k photos). Remember to enable background activity when prompted, or it may be paused in the background. The UI may differ slightly in the latest version.
Want searches like "beach", "night scene", or "family picnic" to be sharper? Run Image Understanding to generate scene, object, and activity tags for every photo. ~24h / 10k, best overnight while charging. Pause anytime — already-processed photos won't repeat.
AI chat runs on a cloud-based model for the best experience. Sign in with email for a higher free quota (guests get less). A network connection is all you need.
After scanning, people are detected automatically. Name each person and set relationships (family / friends / colleagues…), then add facts in AI Memory — PoLang gets smarter in every conversation.
Connect Feishu / Telegram to drive your gallery from IM (e.g. "send me yesterday's landscapes") — results come right back to the chat. Telegram uses long-polling, no public IP needed.
Just say something to the AI
"Find last summer's beach photos"
🗂️ Matching photos, grouped smartly by time — all search, face, and scene recognition happens on-device; photos are never uploaded.
