← All playground

AI Glasses

Wearable Multimodal AI

Creative Technology · Hardware · AI

CreditPersonal build

CreditPublished on Hackster.io

HardwareMini Camera + Video Encoder (RTSP)

AIGoogle Gemini API

RoleMaker & Creative Technologist

DisciplineAIInteractive

Personal build · Wearable multimodal AI

Ask your glasses
what you're looking at.

AI Glasses is a scrappy DIY wearable: a tiny camera clipped to an ordinary pair of frames streams my point of view to a multimodal AI, so I can ask about whatever is in front of me — read it, translate it, solve it — hands-free. No custom silicon, no fancy SDK: just a mini analog cam, a video encoder, an RTSP stream, and the Google Gemini API doing the thinking.

CaptureMini Cam

EncodeVideo Encoder

TransportRTSP Stream

OrchestrateAI Agent

ReasonGemini

GroundRAG

View the full build on Hackster →

The idea

Smart glasses, the scrappy way.

Consumer AI glasses are expensive and closed. I wanted to feel the idea for myself with parts already on my desk — so I clipped an analog mini camera to my everyday frames, piped its feed out as a network video stream, and wired that into a multimodal model. It's admittedly not the elegant way to do it. But it answers the only question that matters at the prototype stage: what does it feel like to look at something and just ask?

Not the most elegant build — just the most fun to test.

How it works

From my eyes to the model.

The pipeline is deliberately simple. The mini cam captures whatever I'm looking at; a video encoder turns that analog feed into an RTSP stream on the network; an AI agent grabs a frame plus my question and hands both to Gemini, whose multimodal models read images, text and audio and answer in plain language. A RAG layer can ground those answers in my own reference data when the question needs it.

RTSP streamingMultimodal promptGemini APIRAG grounding

Architecture diagram — mini cam to video encoder to RTSP stream to AI agent to Gemini and RAG — Schematic flow — Mini Cam → Encoder → RTSP → AI Agent → Gemini + RAG

The hardware

Two off-the-shelf parts.

The whole rig is a pair of glasses, a thumb-sized analog camera, and a small video encoder — that's it. The cam clips to the frame and rides on your temple; the encoder does the unglamorous job of getting the picture onto the network. Cheap, hackable, and entirely reversible.

The build laid out — glasses, mini camera and video encoder — The full bill of materials — frames, mini cam, video encoder

The thumb-sized analog mini camera — Mini cam — thumb-sized, clips to the frame

The video encoder that turns the feed into an RTSP stream — Video encoder — analog feed to RTSP stream

What it can do

Look, ask, answer.

Three quick tests show the range once your view becomes a prompt — the same wearable handling recognition, language and reasoning with nothing but a question.

Demo 01Object detectionPoint at anything — say, a banknote — and ask "what is this?" Gemini names it from the live frame.

Demo 02Language translationLook at text in another language and get it translated on the spot — signage, menus, a line of Chinese.

Demo 03Math problemGlance at a written word problem and the glasses read it, reason through it, and hand back the answer.

gemini_request.py

# send a frame from the RTSP stream + a question to Gemini
from google import genai

client = genai.Client(api_key="YOUR_API_KEY")

response = client.models.generate_content(
    model="gemini-2.5-flash",
    contents=[frame, "What am I looking at?"],
)
print(response.text)

A weekend experiment more than a product — but a real one you can wear, and a hands-on way to feel where wearable, always-available AI is heading. Full write-up and demos on Hackster →