PixelOkay
● On-device editor
Tools
PIXELOKAY VIDEO LAB ✦ CREATOR STUDIO

AI Auto Zoom & Focus

Guide attention to what matters. Detect movement and visual detail, refine the camera path, then export a smooth, focused video—all in your browser.

✦ Automatic focus◎ Editable keyframes↗ Smooth zoom transitions◉ No paid API
YOUR CREATIVE WORKSPACE

Focus studio

Import, analyze, adjust, and export without leaving this page.

Live camera preview LIVE

A focused look at your source footage.

No video loaded
▣⌖✦

Every detail deserves attention.

Choose a video or open the built-in demo. You can automatically detect activity or place your own focus points.

Camera path

Choose a focus point to edit, or click along the timeline to seek.

0 keyframes
Your focus points will appear here.
Focus: 50%, 50%Zoom: 1.0×
↧

Export focused video

Records the edited composition in real time. The file format depends on your browser.

Export rangeImport a video to choose a segment.

Choose up to 90 seconds; focus keyframes remain on the original video timeline.

Format detected at export

Export any selected segment up to 90 seconds, even from a longer source. The browser may support WebM, MP4, or neither.

◎

Stay in control

Every automatic focus point can be selected, moved, adjusted or removed.

⌁

Smooth camera motion

Keyframes use eased interpolation and restrained tracking to avoid abrupt jumps.

◈

Your footage stays local

Analysis and export run on-device. Optional AI models download from third-party hosting.

Ready. Choose a video or load the demo.
HOW IT WORKS

Make screen recordings easier to follow.

How does automatic zoom work?

PixelOkay samples video frames, finds regions of movement or detailed visual activity, then generates camera keyframes. This fast local detector is heuristic, not a semantic AI understanding of every screen action.

Can it recognize objects?

An experimental free on-device object model can be selected. It needs a model download, may be slow, and can fail depending on device support. Your manual camera tools remain available.

Which video formats can I export?

Browser MediaRecorder support determines the format and codec. WebM is commonly supported in Chromium; MP4 is offered only when the browser reports support. Export runs in real time and requires keeping the page active.

Is footage uploaded?

No. Video frames, keyframes and exports remain within the browser. Loading the optional object model contacts its third-party host to download code and weights, not to upload your video.

LOCAL AI SETUP

Set up AI on this device

First, a model needs to be loaded on your device

This optional experimental object detector runs inside your browser. It can download model weights (~43 MB quantized ONNX file plus supporting assets). Your video is not uploaded for inference.

  1. 1. Check device
  2. 2. Download / cache
  3. 3. Set up object AI
  4. 4. Analyze video
Object detection on sampled video frames. Transformers.js 3.8.1, ONNX q8, WebAssembly. Accuracy and speed depend on footage and device.

Device & model resources

Estimates, not live memory
Device RAM / model memoryChecking…

Model memory usage is not exposed.

Browser website storageChecking…

Site storage is not RAM.

ONE ACTION · ALL RECOGNIZED AI MODEL DOWNLOADS

Free AI model storage across PixelOkay

Scan this website origin for recognized WebLLM, Transformers.js and other identifiable model caches from all PixelOkay tools on this exact origin. Other website data and databases are never automatically cleared.

Scan for safely recognized model caches before deletion.

Unrecognized Cache Storage, IndexedDB and OPFS are not removed.

Having trouble downloading or loading AI?

Try Refresh Storage & Cache, a stable network and Retry Setup. Cache cleanup is optional and cannot fix every GPU, CORS, network or worker failure. It can require other PixelOkay tools to re-download models.

Check device information or continue without object detection.
Currently loadedNone

Remove downloaded AI model caches?

Using AI Auto Zoom & Focus

Import a video or sample, choose analysis, review and correct focus keyframes, set the range and record.

Object mode downloads external runtime/model assets and remains experimental. Conventional modes are heuristics. Sampled fingerprints are not authentication. Projects omit video; imports are bounded to 180 seconds and exports to 90 seconds. Real-time codecs, frame timing and audio sync depend on the browser.

Review focus keyframes before recording a video section

Import a supported video, choose conventional motion, text-like or saliency analysis, or request the optional experimental object detector. Inspect the suggested focus path, correct it with manual keyframes, choose a section and record the reviewed result.

Example workflow

  1. Open a short screen demonstration and select the analysis mode that fits its content.
  2. Check each proposed focus point against the frame, adjust the keyframes and reanalyze only when useful; manual keyframes are protected during reanalysis.
  3. Set a short export range, preview the crop and record it in real time. Save project JSON if you need analysis settings and keyframes later.

Authored example: For a tutorial that moves from a settings menu to a chart, check that the zoom follows the active area without hiding the chart labels. Add a manual keyframe at the transition if the suggested path drifts.

Limits and model handling: Motion, text-like and saliency modes are heuristics. Optional object detection downloads third-party runtime/model assets and remains experimental. Imports are limited to 180 seconds and exports to 90 seconds; project JSON omits the source video. Sampled project fingerprints are not authentication. Real-time codec support, frame timing and audio synchronization depend on the browser.

AI Auto Zoom & Focus questions

Will analysis always track the right person or object?

No. Conventional modes are heuristics and optional object detection is experimental. Review and edit the keyframes against the video.

Can a project reopen without the source video?

No. Project JSON stores supported settings and keyframes, not the video. Keep the original and reattach it when reopening the project.

Related tools