3.8 / 5

Scene Describer

Camera describes surroundings in real-time for blind and low-vision users

Serious

Privacy

Offline

Real-time

The Problem

Blind users can't perceive their physical environment; existing tools require photos and cloud processing

Current Solutions (Not Great)

Be My Eyes (requires volunteer), Seeing AI (cloud-dependent), white cane (no detail), guide dogs (expensive, limited)

Who Needs This

Blind and low-vision individuals, their families and caregivers, orientation and mobility specialists

Screen readers handle text but can't tell you what's in front of you. Scene Describer uses your phone's camera to provide continuous audio descriptions of your environment: 'You're in a coffee shop. Counter is 10 feet ahead, slightly to your left. Two people in line. Empty table to your right.' It prioritizes information that matters for navigation and safety—obstacles, people, doors, stairs—and can answer questions about what it sees. Because it runs locally, it works offline and your visual world isn't streamed to any server.

Honest Take

You're competing with Be My Eyes and Microsoft's Seeing AI, which are well-established and backed by serious funding. The technical challenge of real-time vision processing with accurate spatial understanding is brutal—getting distance and direction right is way harder than just describing what's in the frame. That said, the privacy angle and offline capability are genuine differentiators that the cloud-based tools can't match.

Monetization Ideas

Ways to turn this into revenue

Freemium

Free basic, $5-15/mo for pro

Subscription

$5-29/month or $49-199/year

Donations / Tip Jar

$5-50/month from grateful users

Features

Key features that make this app valuable

Continuous scene description
Obstacle and hazard alerts (priority)
Navigation-relevant information first
Question answering ('What's to my left?')
Text reading (signs, menus, labels)
Face detection (without identification)
Configurable verbosity
Spatial audio cues for directions

Build Prompt

Use this prompt with an AI assistant to start building

Build a React PWA called 'Scene Describer' using WebInfer with vision capabilities. UI: minimal—large start/stop button, settings gear, mostly audio-driven. Use getUserMedia({ video: { facingMode: 'environment' } }) for rear camera. Continuous frame capture (2-3 fps for battery). For each frame, use vision model with generateObject to return { priority: 'urgent'|'navigation'|'context', obstacles: [{ type: string, distance: 'near'|'medium'|'far', direction: 'left'|'center'|'right' }], people: { count: number, positions: string[] }, navigation: { pathClear: boolean, doors: string[], stairs: string[] }, text: string[], sceneDescription: string }. Narration priority: urgent obstacles immediately (with vibration), navigation on request or change, context when nothing urgent. speechSynthesis with varying urgency in tone. 'What's ahead?' voice command via SpeechRecognition. Settings: verbosity level, description frequency, voice speed. Completely offline-capable after model load. Test with actual blind users for feedback.

Score Breakdown

Pain