Skip to main content
Preparing workspace

Photo to Art Style

Restyle a photo with TensorFlow.js neural style transfer. Pick a preset or upload a style image, set the strength, and download a PNG or JPG.

Still loading. If the tool doesn’t appear, reload this page.

About this tool

Photo to Art Style

Restyle a photo as artwork with neural style transfer in your browser. Blend custom style images or built-in art styles with adjustable strength, original color preservation, and PNG/JPG download.

Photo to Art Style brings arbitrary neural style transfer directly into your browser via TensorFlow.js. A pretrained MobileNet style encoder extracts a 100-dimensional artistic signature from any reference painting or built-in preset, blends it with your photo's identity vector at your chosen strength (0–100%), and runs a separable transformer network locally on your GPU (WebGL) or CPU. The model, about 12 MB, downloads from our server the first time you add a photo; it is Magenta arbitrary image stylization, ported to TensorFlow.js by Reiichiro Nakano under the Apache-2.0 licence. Style vectors are cached and content images are scaled to 384 px, 512 px, or 768 px, so changing strength or presets re-runs only the transformer. Your photos are not uploaded.

How to use Photo to Art Style

  1. 1

    Upload or paste a content photo (JPG, PNG, or WEBP up to 15 MB) into the Content Photo panel, or pick a built-in sample scene to start immediately.

  2. 2

    Choose one of the 6 built-in painterly presets (such as Starry Swirl, Cubist Mosaic, or Watercolor Garden) or upload your own artwork into the Style Reference panel.

  3. 3

    Drag the Style Strength slider between 0% and 100%, switch output resolution between 384 px, 512 px, and 768 px, or toggle Preserve Original Colors. With Auto-Apply on, the artwork updates after each change.

  4. 4

    Inspect the transformation using Result, Split Compare, or Side-by-Side view, then download the stylized artwork as PNG or JPG or copy it to your clipboard.

Why use Photo to Art Style

  • Strength slider with cached vectors

    Style and content bottleneck vectors are cached in memory, so dragging Style Strength from 0% to 100% only re-runs the lightweight transformer pass, which renders faster.

  • Built-in presets plus any custom style image

    Use any painting, illustration, or texture from your device—or click one of the 6 built-in procedural art styles—to restyle your photo quickly.

  • Browser-side inference

    The model downloads from our server once and runs locally via WebGL, with a CPU fallback. Your photos stay on your device.

Common use cases

  • Turning travel landscapes, architecture shots, or pet portraits into oil-painted, watercolor, or woodblock-inspired wall art without installing desktop software.
  • Testing how different painting textures, brushstrokes, and color palettes transform a photograph before committing to an illustration direction.
  • Applying artistic brushwork to a portrait while toggling Preserve Original Colors so skin tones and natural lighting stay recognizable.

Frequently asked questions

How does browser neural style transfer work?

The tool runs two pretrained TensorFlow.js graph models: a MobileNet-based style predictor that converts a style image into a 100-number artistic embedding, and a fast separable style-transfer transformer that repaints your content photo using that embedding.

Why do the AI models take a few seconds to load the first time?

The first time you add a photo, your browser downloads roughly 12 MB of pretrained TensorFlow.js weights (six files) from our own server route. The model credit: Magenta arbitrary image stylization, TensorFlow.js port by Reiichiro Nakano, Apache-2.0. Later runs reuse the loaded model.

What does Preserve Original Colors do?

Standard neural style transfer borrows both the brushstrokes and the color palette from the style image. Enabling Preserve Original Colors keeps the stylized luminance and brushwork while restoring your photo's original hue and saturation.

Which resolution mode should I choose?

Balanced (512 px) is the default and a reasonable balance of detail and speed. Use Fast (384 px) for quicker updates on older devices, or HD (768 px) for sharper final PNG/JPG exports.

Are my uploaded photos sent to a server?

No. Both images stay in your browser tab's memory and canvas elements. Only the model files are downloaded; your images are not uploaded.