Pick a photo and a style image, such as a painting, a fabric pattern or one of the built-in artworks, and a neural network repaints your photo with that style's colors, strokes and textures. It runs on your own device with TensorFlow.js, with a strength slider, original-color option and before/after comparison.
Last updated:
◂ ▸
PhotoStyled
Working…
photofull style
smoothheavy
Advertisement
Advertisement
How to use the AI Style Transfer
Add your photo by dropping it on the first box, choosing it or pasting it. A sample photo is loaded to start with.
Choose a style: click one of the built-in styles or upload your own style image, for example a photo of a painting, a mosaic or a fabric.
Click Apply style. The first time, the AI model downloads and is cached; then your photo is repainted in a few seconds.
Fine-tune with Style strength, Brush scale and Keep the photo's original colors, and drag across the picture to compare.
Download the result as PNG or JPG.
How fast arbitrary style transfer works
The original neural style transfer (Gatys, Ecker and Bethge, 2015) optimized a picture for minutes until its deep features matched the content of one image and the texture statistics of another. Later work trained feed-forward networks that stylize in one pass, first one network per style, then networks that accept any style. This tool uses that last approach:
Style prediction. A MobileNet-based network looks at the style image (resized by the Brush scale setting) and outputs a 100-dimensional style vector.
Strength mix. The same network also reads your photo. The final vector is strength × style + (1 − strength) × photo, which gives a natural-looking partial effect.
Stylization. A transformer network with conditional instance normalization repaints the photo using that vector, at the AI working size you selected.
Original colors (optional). The brightness of the result is combined with the color of your photo, keeping only strokes and texture.
Settings guide
Setting
Lower
Higher
Style strength
Subtle, the photo dominates
Full style, strong strokes and palette
Brush scale
Large strokes: style image analyzed small (192 px)
Fine strokes: style image analyzed larger (384 px)
AI working size
Faster, less detail (384 px)
More detail, needs more memory (1024 px)
Texture (quick mode)
Smooth painted look
Heavy texture copied from the style image
Tips: portraits look best with lower strength and original colors kept, landscapes and architecture handle full strength well, and a style image with clear, repeating texture gives more convincing results than a detailed scene.
Frequently asked questions
What is AI style transfer?
Style transfer is a technique that repaints one image (the content) using the visual style of another (the style): its colors, brush strokes and textures. A neural network separates what is in the photo, such as shapes and objects, from how a picture looks, and combines the content of your photo with the look of the style image.
How does arbitrary style transfer work here?
The tool uses the fast arbitrary style transfer method of Ghiasi and colleagues (2017) in a version distilled for browsers. A style-prediction network turns the style image into a short list of 100 numbers, a style vector. A transformer network then paints your photo conditioned on that vector in a single pass, which is why any image can be used as a style and results take seconds instead of minutes.
What does the style strength slider do?
In AI mode it mixes the style vector of the style image with the style vector of your own photo before painting, so lower values give a subtler effect that still looks painted. In quick mode it blends the result with the original photo.
Which style images work best?
Images with a strong, consistent look: paintings with visible brush strokes, mosaics, stained glass, woodcuts, fabrics and abstract patterns. Crop away frames, walls and museum labels. Use the Brush scale setting to make strokes bigger or finer, and only use artworks you are allowed to use, such as your own work or public-domain paintings.
Why is the first run slow?
The browser downloads TensorFlow.js and the two style networks the first time, then caches them. After that, a 512 px image usually takes a few seconds on a computer with a graphics card. Larger working sizes need more graphics memory and time.
What is the quick color and texture mode?
It is a classic image filter, not AI. A Kuwahara filter flattens the photo into painted patches, the colors are matched to the style image in the Lab color space (Reinhard color transfer), and a high-pass texture from the style image is layered on top. It is instant and works offline, but it cannot reproduce brush strokes or shapes like the neural network.
Are my photos uploaded?
No. Both the photo and the style image are processed on your device. Only the model files are downloaded from a public CDN; nothing you add is sent anywhere.
Are all the tools on FreeToolCabin tested for quality and accuracy?
Yes! Every tool on our site undergoes automated Q/A testing in real browser environments to ensure high precision, correct output, and zero errors. If you happen to spot a bug or have a suggestion, please contact us at [email protected] and our team will review it right away.