cd /news/computer-vision/aethercut-hardware-acceleration · home topics computer-vision article
[ARTICLE · art-46848] src=dev.to ↗ pub= topic=computer-vision verified=true sentiment=↑ positive

AetherCut Hardware acceleration.

AetherCut implements hardware-accelerated video editing entirely in the browser, using WebGL for real-time chroma key compositing and WebAssembly to run OpenAI's Whisper model locally for offline transcription in 14 languages. The 'prefer-hardware' flag delegates encoding to the GPU, and the WebGL fragment shader performs green screen removal on the client side without any network requests.

read1 min views1 publishedJul 1, 2026

The hardware Acceleration: 'prefer-hardware' flag tells the browser to delegate to the GPU encoder first. On Chrome 94+ and Edge, this is widely supported across Windows, macOS, and Android.

WebGL — Chroma Key Without a Server

Green screen removal is computationally expensive if done naively. AetherCut handles it via a WebGL fragment shader that runs on the GPU — no frame-by-frame CPU pass, no upload to an AI service.

The shader samples each pixel's hue and saturation, compares it against the chroma key target color, and writes a transparency mask in real time. This runs at frame rate during preview.

// Fragment shader excerpt — chroma key precision mediump float;

uniform sampler2D uTexture;

uniform vec3 uKeyColor;

uniform float uThreshold;

varying vec2 vTexCoord;

void main() {

vec4 color = texture2D(uTexture, vTexCoord);

float diff = distance(color.rgb, uKeyColor);

float alpha = diff < uThreshold ? 0.0 : 1.0;

gl_FragColor = vec4(color.rgb, alpha);

}

Because this executes entirely on your local GPU, webcam footage processed through Green Screen never generates an outbound network request.

Whisper — Offline Transcription in 14 Languages

Transcription is where most "AI" editors quietly phone home. They send your audio to a cloud ASR service and return a transcript. You never see the request because it's abstracted away.

AetherCut runs the Whisper model locally via WebAssembly. The model weights download once and are cached. After that, transcription runs entirely on your device — word-level accuracy, 14 languages supported, no audio ever transmitted.

── more in #computer-vision 4 stories · sorted by recency
── more on @aethercut 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/aethercut-hardware-a…] indexed:0 read:1min 2026-07-01 ·