style transfer

basicwidespread

An AI method that preserves the content of an image, text, or audio but replaces its appearance with the style of another work or creator.

In plain terms

Imagine giving your vacation photo to a painter and asking them to paint the same scene, but as Van Gogh would have done. The houses, trees, and people remain in the same places, only the brushstrokes, colors, and mood change. Style transfer does exactly the same, but with a machine instead of a painter, typically within a minute.

The essence is separation: the system handles separately what is in the image (content) and how it is presented (style). That is why the same photo can become a watercolor, comic, cartoon frame, or oil painting.

It does not work only on images. With text, it rewrites the same message from formal to friendly or from technical to accessible; with audio, it reproduces a melody with different instrumentation or a sentence with a different voice tone.

A traditional photo filter modifies the image according to a pre-written rule, typically affecting colors, brightness, or sharpness. Style transfer also redraws the structure of the image, meaning it recreates lines, textures, and shapes.

How it works

The system receives two inputs: the content (for example, your photo) and a style reference, which can be a specific image or a text instruction, such as "in watercolor style".
A neural network analyzes the content and highlights its structure: what is where, and what shapes and outlines compose it.
It extracts characteristic patterns from the style sample: color palette, brushwork, texture, line weight (or, for text, vocabulary, sentence rhythm, and tone).
It then produces a new image that resembles the content in structure and the style in patterns, seeking a balance between the two.
Modern image generators no longer perform this as a separate procedure: they interpret the uploaded image and the requested style together and essentially reconstruct the image.

An everyday example

You upload a family photo to an image generator and ask it to turn it into a Japanese anime style. After a short wait, you get the same scene as a drawing; social media filled with such images in the spring of 2025. The same happens when you ask a chatbot to rewrite your angry email as a polite formal letter.

Who is affected

everyoneIt is now accessible with a single sentence in mobile photo editors and chatbots, so you encounter the results daily on social media.

marketerIt lets you quickly create images with a consistent visual identity and text variants in different tones, but the legal background of the style used must be checked. → Marketing

parent, student, teacherA good illustrative tool for art history or essay teaching, yet it can easily overshadow the student's own voice. → Education & Family

lawyer, citizenThe machine imitation of living creators' styles raises open copyright and ethical questions. → Law

A common misconception

MythStyle transfer is just a flashier photo filter laid over the image.

In realityThe system does not recolor the image but redraws it, so details may change, disappear, or new ones may appear. Technically, you receive a newly generated image. Legally, however, this counts as a modification of the original image, so you cannot freely use someone else's photo or artwork just because you restyled it.

Limits and risks

Details can be damaged: faces, hands, and text often distort, and the redrawn face does not always resemble the original.
Style itself is generally not protected by copyright, but the commercial imitation of a living creator's recognizable signature can trigger legal and ethical disputes.
With text, changing the tone can shift the meaning: rewriting may omit nuances or emphasize different aspects, so always review it.
If you upload your own or others' photos to an online service, check what the provider can do with the image and whether you have consent from the people involved.

Similar, but not the same

image generation – Image generation creates a new image from a text description, while style transfer restyles existing content.
deepfake – A deepfake impersonates a real person with deceptive authenticity. Style transfer changes the manner of appearance and does not claim that something actually happened.
photo filter – A filter is a pre-written modification (for example, color, brightness, sharpness), while style transfer reconstructs the entire image based on learned patterns.

Related concepts

Broader

generative AI, generative media, machine learning

Narrower

neural style transfer, text style transfer

On the edge

image generation, diffusion model, deepfake, voice cloning, Neural network, prompt

A brief history

Neural style transfer was launched by a 2015 study by Leon Gatys, Alexander Ecker, and Matthias Bethge, which showed that a neural network can separate and remix the content and style of an image. In 2016, the Prisma mobile app brought the method to the masses.

Where things stand

As of late 2026, style transfer is a built-in, everyday feature of image generators, photo editors and chatbots, for images, text and increasingly video. Today the debate over whose style may be imitated is sharper than the technical questions.

Status: 2026 November

Also known as

style transfer · style transfer · style transfer · neural style transfer · style transfer · neural style transfer · NST · artistic style transfer

Sources

Gatys, Ecker, Bethge: A Neural Algorithm of Artistic Style (arXiv, 2015)

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.