# AI image tools that run in your browser

Browser tools powered by Apple vision models: depth map generator, zero-shot image classifier, image–text similarity, image Q&A and a device benchmark. Images never leave your device.

Canonical URL: https://aiidelist.com/blog/apple-ai-browser-tools

Language: en

Published: 2026-09-24

Updated: 2026-09-24

TOOLS · RUNS ON YOUR DEVICE

Each tool downloads an open model once and then runs on your GPU through WebGPU (or WebAssembly where noted). Your images, labels and results stay on this device.

 [In-browser

## Image Q&A

Ask questions about a photo or screenshot, read text, explain a chart.

**Powered by**

FastVLM-0.5B

**Download**

≈1.1 GB

Open tool](/fastvlm/demo#image-playground) ·  [In-browser

## Depth Map Generator

Convert one image into a depth map and download it as PNG.

**Powered by**

Depth Pro · Depth Anything V2

**Download**

50 MB – 600 MB

Open tool](/blog/depth-map-generator) ·  [In-browser

## Zero-Shot Image Classifier

Type your own labels and rank them against an image. No training.

**Powered by**

MobileCLIP-S0

**Download**

≈55 MB – 110 MB

Open tool](/blog/image-classifier) ·  [In-browser

## Image–Text Similarity

Score how well sentences describe an image with CLIP embeddings.

**Powered by**

MobileCLIP-S0

**Download**

≈55 MB – 110 MB

Open tool](/blog/image-text-similarity) ·  [Local guide

## Image to 3D

Turn one photo into 3D Gaussians for nearby views, or a relightable asset.

**Powered by**

SHARP · Luce

**Download**

Local install

Open tool](/blog/apple-sharp-image-to-3d) ·  [In-browser

## Benchmark Your Device

Measure model load, first token and tokens per second on your GPU, then share the result.

**Powered by**

FastVLM-0.5B

**Download**

≈1.1 GB

Open tool](/blog/benchmark-fastvlm-device)

## How these tools work

- Models are fetched from Hugging Face on first use and cached by your browser.
- Inference runs inside a Web Worker on your GPU; FastVLM.net never receives your images.
- Usage events record only fixed step names, the model and timings—never content.

## Want the model, not just the tool?

Every tool links to the Apple model behind it, with Python, MLX and browser instructions.

[Browse Apple models](/blog/apple-ai-models)