H 298 / 3212010s

Neural Style Transfer

2015, Leon Gatys's algorithm

Typeface · Henny Penny

Seated Portrait with Transferred Brushstrokes. A seated man's recognizable photographic portrait is overlaid with orange, gold, and green directional brushstrokes.
Seated Portrait with Transferred Brushstrokes. A seated man's recognizable photographic portrait is overlaid with orange, gold, and green directional brushstrokes.Kira Wales · CC BY-SA 3.0 · via Wikimedia CommonsView source ↗

Prompt

Apply a Neural Style Transfer effect, translating photographic surfaces into layered painterly brushstrokes, swirling pigment textures, and harmonized teal, ochre, and crimson tones. Let repeated stroke patterns follow major forms, preserving recognizable contours while replacing fine photographic detail with expressive painted texture. Keep the original subject and composition.

Identify

You see a photo altered to look like a painting. It keeps the original shapes but adopts another image's colors and brushstrokes.

Principle

An artificial neural network1 separates an image's content from its style. It then combines the content mathematically with a new style.

Example

Leon Gatys, Alexander Ecker, and Matthias Bethge: A Neural Algorithm of Artistic Style, 2015.

Try it out

Squint at a style-transferred image. Notice how the broad structural outlines remain recognizable, but the surface texture completely changes.

Background

In 2015, researchers led by Leon Gatys published a breakthrough paper on image processing. They used convolutional neural networks2, originally built to recognize objects in photos. The team discovered that different layers of the network process different image traits. Deep layers capture the spatial layout3, or content. Shallow layers capture texture, color, and patterns, defining the style. By running a photo and a painting through the network, they extracted these elements. An algorithm then merged the photo's content with the painting's style. This digital technique sparked a massive wave of artificial intelligence art.

Distinction

Unlike simple digital photo filters, it does not just overlay a color tint or blur. It completely reconstructs the image using mathematically learned textures.

Aftermath

It reveals that artificial intelligence can mathematically isolate and manipulate the abstract concept of artistic style.

Further examples · 7 images

Profile Portrait with Flowing Color Patterns
02 Profile Portrait with Flowing Color PatternsA woman in profile wears glasses and a tall headwrap, with multicolored, flowing textures extending across her face, clothing, and background. The preserved portrait structure beneath a pervasive painterly pattern clearly illustrates the visual combination of photographic content and transferred style.Nettrice Gaskins · CC BY-SA 4.0 · Wikimedia Commons
AI illustration: Triptych of bridge photograph and Van Gogh style transfer
03 AI illustration: Triptych of bridge photograph and Van Gogh style transferA triptych shows a photograph of a bridge, a swirling blue pattern, and a final image combining both. This sequence demonstrates how a neural network maps the brushwork of one source onto the geometry of another.Illustration generated for the Atlas (AI)
AI illustration: Monitor showing neural style transfer of fruit bowl
04 AI illustration: Monitor showing neural style transfer of fruit bowlA computer monitor displays a bowl of pears alongside a thick impasto painting and the resulting stylized output. The output retains the photographic placement of the fruit while adopting the color palette of the painting.Illustration generated for the Atlas (AI)
AI illustration: Watercolor neural style transfer of a city street
05 AI illustration: Watercolor neural style transfer of a city streetA series of prints displays a cyclist in London transformed into a soft watercolor composition. The process successfully extracts the textures and hues of the watercolor swatch to redefine the photographic scene.Illustration generated for the Atlas (AI)
AI illustration: Gallery display of egret image style transfer
06 AI illustration: Gallery display of egret image style transferThree canvases on a gallery wall show a white egret, an abstract blocky pattern, and a synthesis of the two. The final large-scale piece illustrates the algorithmic reinterpretation of a wildlife photo into abstract geometric forms.Illustration generated for the Atlas (AI)
AI illustration: Spiral staircase transformed via neural style transfer
07 AI illustration: Spiral staircase transformed via neural style transferA dual-monitor setup shows a spiral staircase transformed into a textured work of pink and blue oil paint. The intricate architectural lines of the staircase are preserved through the application of heavy digital brushstrokes.Illustration generated for the Atlas (AI)

Question

Why do some style transfers look messy or unrecognizable?

Show possible answer +

The algorithm might weigh the style too heavily. This causes the painting's textures to overwhelm the original photo's structural content.

Sources