AI Tools
Remove backgrounds, upscale photos, transcribe speech and more with AI models that run privately on your own device.
Most AI media tools send your photos and recordings to a cloud GPU. These tools take a different approach: the model is downloaded to your browser once, cached, and then run locally with WebGPU or WebAssembly. Your images, audio and video never leave your device.
This category is being built now. Each page explains what the tool will do, which open models it is based on, and what to expect from on-device performance compared with cloud services.
Image AI
5 toolsVideo AI
6 toolsAudio & Speech AI
6 toolsSpeech to Text
Transcribe audio and video privately.
Text to Speech
Natural voices, generated on-device.
Vocal Remover
Split vocals and instruments.
AI Noise Suppression
Clean up voice recordings.
Filler Word Remover
Cut “um” and “uh” automatically.
Subtitle Translator
Translate subtitles on-device.
Smart Tools
5 toolsGuide
How on-device AI works in the browser
Libraries such as ONNX Runtime Web, Transformers.js and MediaPipe run neural networks directly in the page. WebGPU gives access to your graphics card for fast inference. Where WebGPU isn't available, a WebAssembly fallback runs on the CPU. It is slower but works almost everywhere.
The trade-off is a one-time download. Small models are a few megabytes, while speech recognition and image restoration models can be tens to hundreds of megabytes. Once cached by the browser, they load quickly on later visits and can work offline.
Guide
What to expect from small models
On-device models are smaller than the largest cloud models, so results can be slightly less polished on difficult inputs such as fine hair, heavy noise or strong accents. For everyday photos, recordings and documents, modern compact models are accurate enough, and the privacy benefit is significant.
| Backend | Runs on | Relative speed | Availability |
|---|---|---|---|
| WebGPU | Graphics card | Fastest — often several times faster than CPU | Recent Chrome and Edge; expanding to other browsers |
| WebAssembly (SIMD + threads) | CPU | Moderate | All modern browsers |
| WebAssembly (basic) | CPU, single thread | Slowest | Universal fallback |
Frequently asked questions
Common questions about ai tools.
Why run AI in the browser instead of the cloud?
Privacy and cost. Your files never leave your device, there is no upload wait, and because no server GPU is involved, the tools can be offered for free.
How big are the model downloads?
It depends on the task. They range from a few MB for face detection to tens or hundreds of MB for speech recognition or image restoration. Each model downloads once and is then cached by your browser.
Will it work on my phone?
Lighter models such as face detection and OCR run well on recent phones. Heavy tasks such as upscaling, vocal separation or long transcriptions are best on a laptop or desktop with a decent GPU.
When will the AI tools be available?
They are in development and will roll out one at a time. Each tool page will switch from a guide to a working tool when it launches.