Decoder. plain-English AI glossary

Style Transfer

▼ Fading

Applying the artistic style of one image to the content of another - turning a photo into a painting in a chosen style.

Think of it like

Using Photoshop filters or art classes: the content (what) stays the same, but the presentation (how its painted) changes.

Example

Photo of a dog -> painted in Van Gogh stylized starry night style. A landscape photo -> watercolor sketch.

How it actually works

Classic approach (Gatys et al.): Gram matrix matches (texture statistics) swap style while preserving content. Defined via feature activations of a pretrained CNN (VGG). Slow (per-image optimization) but artistic. Modern alternatives: feed-forward networks trained on specific styles (faster but less flexible), or diffusion-based approaches. Perceptual loss guides optimization. Trade-off between content fidelity and style strength.

For product teams

Dated by generative models but still useful for targeted artistic effects.

For engineers

Classic method: Gram loss + content loss via perceptual features. Modern: generative models + prompt engineering.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open DecoderAdd to Chrome