PixelOkay
IMAGE STUDIO
Tools

PIXELOKAY / AI IMAGE TOOLS

AI Image Layer Separator

Turn a flat image into movable object and text cutouts. Separate with free on-device AI, refine masks, and export transparent layers—without uploading your photo.

Private image workspaceImages stay in your browser. The free models download only after you choose to run separation.

Open an image or project

PNG, JPEG, WebP (24 MB max), or a saved PixelOkay layer project.

Separate image elements in three steps

01

Import

Open an image or previous layer project. Everything stays in this browser tab.

02

Extract

Open Model Setup to consent to a free model download, then click Separate layers. Review automatic object masks, detect text regions, or select any rectangle.

03

Edit & save

Reposition, refine, reorder, and hide layers. Export a transparent PNG, all cutouts, or a reusable project.

Using AI Image Layer Separator

Open an image, extract supported masks or make manual selections, then review each raster layer. Refine masks and arrangement before exporting the composite or layer bundle. Save project JSON for continued editing.

Panoptic masks are not universal precise matting. Layers and OCR text regions are raster cutouts, not live fonts or PSD objects; moving them exposes transparent holes. No background inpainting or hidden-surface generation is supplied. Optional runtime/model/OCR assets download only on separation. Source/layer/aggregate-pixel/project budgets apply; JSON embeds image data and Canvas exports drop metadata.

References: Panoptic segmentation model · Tesseract.js

Separate reviewed raster cutouts into movable layers

Open a supported image and choose manual selections or optional panoptic segmentation and OCR cutouts. Refine each mask, arrange the resulting raster layers and export a transparent composite PNG, a layer ZIP or project JSON.

Example workflow

  1. Start with a clear image and decide whether manual selection or an optional model-generated mask is a better first pass.
  2. Inspect mask edges and OCR regions, refine and reposition layers, and check the transparent areas revealed when cutouts move.
  3. Export the composite or layer package; save project JSON if you need the image data and supported edits for continued work.

Authored example: Separate a person and a short headline from a simple event poster. Zoom in around hair and letter edges, refine the masks, and inspect the empty regions that appear when either cutout is moved.

Limits and model handling: Optional DETR/Tesseract runtime and model assets download when separation is requested. Panoptic masks are not precise universal matting; OCR and object regions remain raster cutouts, not live fonts or PSD layers. Moving a cutout reveals transparency; the tool does not inpaint hidden content. Project JSON embeds image data and Canvas output drops source metadata; image, layer and pixel budgets apply.

AI Image Layer Separator questions

Can it rebuild the background behind a moved object?

No. It separates visible pixels into cutouts; moving a layer exposes transparent areas rather than generating hidden background content.

Can I edit the text as live typography afterward?

No. OCR regions are raster cutouts. Keep the source file if you need editable text or typography.

Related tools

LOCAL AI SETUP

Set up AI on this device

First, a model needs to be loaded on your device

AI Image Layer Separator uses the on-device DETR panoptic object-segmentation model via Transformers.js. Runtime and model files download only after you choose Download & Load. Your image stays in this browser tab; model files may download from external hosts.

  1. 1. Check device
  2. 2. Download / cache
  3. 3. Prepare image model
  4. 4. Extract object masks
DETR Panoptic (Xenova/detr-resnet-50-panoptic) · Transformers.js 3.8.1 · WASM/CPU · object segmentation only. AI produces raster object masks, not image inpainting, editable text or PSD vector layers.

Device & model resources

Measurements may be approximate
Device RAM / model memoryChecking…

Live RAM and model working-memory use unavailable.

Checking browser WASM and Worker support…

Browser website storageChecking…

Site-wide estimate includes all same-origin PixelOkay tools, and is not RAM.

ONE ACTION · ALL RECOGNIZED AI MODEL DOWNLOADS

Free AI model storage across PixelOkay

Scan this website origin for WebLLM and Transformers.js model caches and recognizable model files, including other PixelOkay tools. Only positively identified model artifacts in Cache Storage can be removed here, including those cached by other tools. Inaccessible IndexedDB, OPFS, projects, drafts and other user data remain untouched.

Run a scan to identify safely removable model artifacts. Unknown browser-model stores cannot be deleted safely.

Checking model cache does not download files. A complete cache is confirmed only by successfully loading the model; partial entries are not treated as ready.

Automatic panoptic segmentation is optional. Manual rectangles, color selection and raster-layer edits work without the model.

The first object segmentation may take longer than subsequent ones. Live RAM and storage measurements are not the same.

Having trouble downloading or loading AI?

First try Refresh Storage & Cache and Retry Setup. Check internet connectivity, runtime/CDN access and device memory. A failed worker import or unsupported WASM is not necessarily a storage problem.

Do not clear all website data: that can remove saved work across PixelOkay. Only safely identified model assets are considered here.

Technical details and diagnostics

Transformers.js @3.8.1 loads Xenova/detr-resnet-50-panoptic in WASM mode. Its browser Cache API can reuse the exact same artifacts across compatible tools on this origin. Model caches in unknown providers or browser databases are not deleted.
CROSS-TOOL MODEL CACHE CLEANUP

Delete recognized model downloads?

Other PixelOkay AI tools may need to re-download model files. Close active AI tabs first. Unknown caches and IndexedDB are untouched; never clear general website data without exporting your work.