Images, video, voice and music
Style transfer
In short
Style transfer is changing an image so it keeps the content of one picture while adopting the visual style of another, such as brushwork, colour and texture. AI models can do this through optimisation or a trained fast-transfer network.
- Also called
- Neural style transfer, Neural transfer
- Topic
- Images, video, voice and music
- Level
- Beginner
- First described
- Leon Gatys, Alexander Ecker and Matthias Bethge introduced the modern neural style-transfer method in A Neural Algorithm of Artistic Style in 2015.
In the classic method, a content image and a style image are passed through a trained vision network. One set of internal features represents the scene and another represents visual patterns such as texture and colour relationships. The system adjusts an output image to reduce both content and style differences12.
Gatys, Ecker and Bethge introduced this neural approach in 2015. Their paper described separating and recombining content and style using neural representations, allowing an image to retain its subject while taking on the appearance of an artwork1.
Style transfer is not the same as copying a filter or generating a new scene from a text prompt. The classic approach starts with image content and a style reference, and the balance between the two affects the result. Modern systems can train a network to produce stylised images directly, which is faster than optimising every output image2.
You see it in art apps, creative filters and design experiments. TensorFlow's official tutorial demonstrates blending a content image with a style reference, including a fast model that can apply arbitrary styles; PyTorch also provides an implementation using content and style losses2.
Example
Give a photo of a bicycle by a river and a painting with thick blue brushstrokes. Style transfer can preserve the bicycle and river layout while rendering them with the painting's colours and texture.
Why it matters
It shows how an AI system can treat scene content and visual appearance as partly separate signals. That makes it useful for creative editing, while also raising questions about permission when the style reference comes from a living artist.
Questions people ask
Does style transfer copy an artist's whole painting?
Not necessarily. The classic method uses a style image's visual patterns while preserving the content image's subject and layout. The output can still resemble the reference, so permission and copyright questions depend on the images and use12.
What is the difference between style transfer and image generation?
Style transfer normally begins with a content image and a style reference. Image generation can create a new image from text or other conditions without preserving a particular source scene. Some modern tools combine both approaches2.
Is style transfer always slow?
No. The original optimisation-based method updates each output image, but trained fast-transfer networks can apply a style much more quickly. TensorFlow describes both approaches in its tutorial2.