PixelOkay
◉ Photos remain on your device
Tools
PIXELOKAY · IMAGE TOOLS CANVAS EXPANSION

AI Image Extender

Extend scenery beyond the original frame, create space for a new composition, and resize your canvas without stretching the photo. Fully local processing, with optional browser AI.

✓ All four directions✓ Aspect ratio presets✓ Exact original pixels✓ Download locally

YOUR CREATIVE WORKSPACE

AI Image Extender

Free · No account

COMPOSITION PREVIEW

Make space for more.

Your canvas starts here

Upload a photo or try the included landscape to see how expansion works.

No image loaded—

Preview is downscaled. Local fill exports at full requested dimensions; experimental AI extrapolates from a 512×512 working image.

Designed for actual creative work

Preserve your subject

The original image area is placed at exact original pixel size without upscaling or stretching.

Extend selectively

Add space above a skyline, turn landscape photos vertical, or make room for text in social posts.

Work privately

Photos are decoded and edited in your browser; no upload or account is required.

Know the limits

Natural scenery usually extends better than text, hands, faces near edges and precise architecture.

LEARN THE WORKFLOW

How to extend an image without stretching it

01

Add a picture

Choose a locally stored image. The original stays in memory on your device.

02

Set the new composition

Select 16:9, 9:16, square or any other preset, then fine-tune how much to add to each side.

03

Process and download

Select a local fill method or optional experimental AI, press Apply, compare, and export your image.

Frequently asked questions

Does expanding an image stretch the original?

No. The source pixels remain exactly the same dimensions and positions within the new canvas.

Is the free AI extension truly local?

Yes. The LaMa model runs in your browser, but its large weights and ONNX runtime must be downloaded from external hosts. Image pixels are not sent to those hosts. Availability depends on your device, connection, and browser.

Will it perfectly generate new scenes?

No. The local methods extend existing edges. The optional LaMa model is primarily an inpainting model repurposed for outpainting, so large missing areas can be inconsistent. This is not a text-prompt diffusion generator.

LOCAL AI SETUP

Set up AI on this device

First, a model needs to be loaded on your device

Experimental LaMa image extension runs locally. Explicitly load the free model (approximately 208 MB) first. Your photo stays in this browser; model/runtime assets download from external CDNs. Context, mirror and edge modes need no AI.

  1. 1. Check device
  2. 2. Download / cache
  3. 3. Initialize selected AI
  4. 4. Extend image
LaMa FP32 (512×512) · ONNX Runtime Web 1.22.0 · WASM · about 208 MB model download. RAM requirements vary; browsers do not expose live model-memory usage.

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 capabilities, then choose whether to load AI.
Currently loadedNone

Remove downloaded AI model caches?

Using AI Image Extender

Load an image, choose expansion amounts or an exact canvas that contains the whole source, and position the original. Select a local algorithm or experimental LaMa mode, apply, compare and export. Settings JSON contains configuration rather than the source image.

Context/mirror/edge modes are conventional algorithms. Experimental LaMa downloads a large model and runtime and repurposes inpainting for expansion; it is not prompt-directed scene generation. The source is retained without resizing/cropping, within 20 MP/8192-pixel output budgets. V3 also supports optional extension-only color/clone brush repair and portable embedded projects within its 42 MB file cap. PNG can preserve the source RGBA crop; JPEG and WebP may be lossy. Input metadata is not retained.

References: LaMa model and provenance · ONNX Runtime

Workflow and output choices for AI Image Extender

Expand the canvas around a supported image, position the complete source within it and compare conventional context, mirror or edge fill with the optional experimental LaMa mode.

Exported image formats are separate from the settings JSON, which stores configuration rather than the source photo. The source remains unresized and uncropped inside the new canvas within output limits.

Example workflow

  1. Load an image and set expansion amounts or an exact canvas that contains the full source.
  2. Position the source, then compare context, mirror or edge fill before considering experimental LaMa.
  3. Review the whole canvas and export PNG, JPEG or WebP; keep the original image because settings JSON does not include it.

Authored example: Use a fictional landscape and add canvas space above and to one side. Compare context fill with mirror fill and inspect where each repeats or blends pixels; do not treat the result as a newly photographed scene.

Limits and model handling: Context, mirror and edge are conventional algorithms. LaMa is experimental inpainting repurposed for expansion, not prompt-directed scene generation; it downloads a large model and runtime on demand. Canvas, file and output budgets apply.

AI Image Extender questions

Do the local context and mirror modes use an AI model?

No. Those are conventional image-fill algorithms; the optional LaMa mode is the model-based experimental option.

Does the settings JSON contain the original photo?

No. It records configuration rather than the source image. Keep the original separately.

Related tools