How do you use this tool?
- Drag a video onto the drop zone or pick one via the file dialog (MP4, MOV, WebM, up to FullHD 1920×1080, up to 500 MB).
- Pick a model: Quality (best hair edges, requires a modern GPU) or Speed (faster, works on weaker hardware and longer videos).
- Choose the blur intensity — light for a subtle depth-of-field effect, strong for an unrecognizable background.
- Wait for the one-time model download — afterwards it's cached in your browser, available offline.
- Preview the result and download the finished MP4.
Why blur instead of remove?
Removing the background completely makes sense when you want to drop the person onto a fresh backdrop later — green-screen compositing, virtual studio looks. For most everyday recordings that is overkill: a tutorial, a vlog, a Twitch stream feels more natural with the room kept and just softened. You preserve the real lighting, the real depth cues, and you don’t have to composite a new scene.
Cloud-based services for this effect have to upload your video to a server, process it there, then hand you back a download URL. That’s not a design flaw — it’s architecturally unavoidable: GPU inference costs money, which gets recouped via subscriptions, watermarks, or credit limits.
Modern browsers flip this equation. Through the WebGPU interface, the browser talks directly to your device’s GPU and runs the AI model entirely on your machine.
How does video background blur work?
Three modern browser interfaces, all available in current desktop browsers since 2024:
In-browser video decoding: Your video is unpacked frame by frame without an external codec library. The browser uses your GPU’s hardware decoder — the same one used for streaming.
AI person segmentation on the GPU: A specialized neural network for image segmentation runs directly on your GPU. For each frame it computes an alpha matte: a grayscale image telling the compositor how strongly each pixel belongs to the person in the foreground.
Frame composition and encoding: Each frame is rendered with the chosen blur applied to the background; the sharp person is then placed on top through the alpha matte. The result is encoded back to MP4 with H.264 — universally playable.
How do you pick the right blur intensity?
Light (1–6): Subtle depth-of-field, room still recognizable — good for warm, personal recordings where context matters.
Medium (7–13): Classic webcam look, like Zoom or Google Meet — background no longer readable but colors and lighting still come through. Recommended default for tutorials and talking-head videos.
Strong (14–20): Background practically unrecognizable — for recordings where the room is clearly private (bedroom, sensitive documents on the wall, other people in frame).
How do you pick the right model?
Quality uses an advanced segmentation model optimized for fine structures — especially hair edges, fur, and glass. Requires a modern GPU.
Speed uses a lighter portrait model that runs near real-time on modern laptops with WebGPU and works reliably in CPU fallback mode on weaker hardware. Hair-edge quality is slightly below the Quality mode, but for long videos and older hardware it’s the better choice.
If your hardware doesn’t support Quality mode, the tool transparently downgrades to Speed mid-load — without an error mid-processing.
What should you know about privacy and the EU AI Act?
Your video never leaves the browser. No cookie records the file name or resolution. No registration, no email, no account.
In the finished state, the page shows the notice: “This video was edited with AI (background blurred).” This satisfies the labeling requirement of the EU AI Act (Article 50), in force from August 2026 for AI-manipulated media. The notice is informational — the responsibility for visible labeling when republishing rests with you.
What are the limits?
Temporal consistency: The model processes each frame independently. On fast hair movement you may see slight flickering at the edges. A stronger blur intensity usually masks it.
Audio: The first version outputs the video stream only. Audio passthrough is planned for a later version. Workflow: keep the original video open in your editor and place its audio onto the processed clip — frame count and frame rate are identical, so no lip-sync drift.
Hardware: Quality mode requires a modern GPU with WebGPU support. On older devices or iOS Safari the tool falls back to Speed mode automatically.
Resolution: Up to FullHD (1920×1080). File size: Soft limit 500 MB.
Which related video tools help next?
Other tools from the kittokit ecosystem that fit this workflow:
- Video Background Remover — cut the background out completely instead of blurring it, for green-screen-style workflows.
- Webcam Background Blur — the same effect live during a video call, no recording needed.
- HEVC to H.264 — convert iPhone videos (HEVC/MOV) into universally playable H.264 MP4.
Last updated: