Style transfer can make your photo look like Van Gogh painted it, or like you applied a bad Instagram filter. The difference is in the style reference image — here's how to pick one that works.
You upload a photo of your dog and a reference image of Starry Night. You expect a swirling, dreamlike portrait. You get… a blue-tinted mess where the dog's face is barely recognizable and the swirls look like someone smeared Vaseline on the lens. What went wrong?
Our AI style transfer tool applies the artistic style of one image to the content of another. When it works, it is magic. When it does not, it is because the style reference and content image are fundamentally incompatible. Here is how to pick references that actually produce good results.
Style transfer uses a neural network that has already been trained to recognize images. The network has layers: early layers detect edges and textures, middle layers detect shapes and patterns, deep layers detect objects and faces. Style transfer separates "what is in the image" (content, from deep layers) from "how it is painted" (style, from early and middle layers), then recombines the content of your photo with the style of the reference.
This is why the reference image matters so much. The network extracts style from textures, brushstrokes, and color palettes — not from composition or subject matter. A reference with strong, distinctive textures (thick oil paint, visible brushstrokes, high-contrast color blocks) transfers well. A reference with smooth gradients and no visible texture (airbrushed digital art, soft-focus photography) transfers poorly — there is no "style" to extract.
Strong, visible texture: oil paintings with impasto (thick paint), watercolors with visible paper texture, pencil sketches with visible strokes. The network needs texture to grab onto. A perfectly smooth digital illustration transfers as a mild color shift with no stylistic change.
Distinctive color palette: a reference with a narrow, intentional palette (mostly blues and golds, mostly warm earth tones) produces a cohesive result. A reference with every color in the rainbow produces a chaotic result where the colors fight each other.
Moderate complexity: a simple abstract painting transfers as a few color blocks — underwhelming. An incredibly detailed painting with hundreds of tiny elements transfers as visual noise — the content gets lost. Mid-complexity references (a landscape painting, a portrait with visible brushwork) produce the best balance of style and recognizable content.
Matching content type to style type: a portrait photo paired with a portrait painting reference works better than a portrait photo paired with a landscape painting reference. The network tries to match features across the two images, and matching a face to a face gives cleaner results than matching a face to a tree.
Our style transfer tool offers two modes:
Artistic mode: the style is applied strongly across the entire image. Brushstrokes, color palette, and texture are fully transferred. Best for: obvious artistic transformations where you want the result to look like a painting. The content image is still recognizable, but the style dominates.
Realistic mode: the style is applied more subtly, preserving more of the original photo's detail. Best for: subtle enhancements where you want a hint of style without losing photographic realism — making a photo look like it was shot on film stock, or adding a painterly quality to skin tones without distorting facial features.
When to use which: artistic mode for social media, creative projects, and prints where the "wow, that looks like a painting" reaction is the goal. Realistic mode for portraits you still want to look like photographs, product images that need to remain recognizable, and any use case where style should enhance rather than replace the content.
Face distortion: if facial features get smeared or warped, switch to realistic mode and reduce style strength. Faces are the hardest content for style transfer because humans are exquisitely sensitive to facial distortion — a slightly smeared eye ruins the entire image.
Style reference too dark/bright: if the style reference is predominantly dark (a chiaroscuro painting), the result will be dark — your content may become barely visible. Choose references with a brightness range similar to your content photo.
Color clash: if your content photo is warm (sunset, indoor lighting) and your style reference is cool (winter scene, blue-period painting), the result will have a muddy color temperature. Match warm content with warm style references, cool with cool.
For generating images from scratch rather than transforming existing ones, see our AI image generator. And for a step-by-step walkthrough, read our style transfer step-by-step guide.
Style Transfer
Apply artistic styles to your photos using AI.
AI Image Generator
Turn text into stunning AI images with SDXL. No watermark, instant download in JPG, PNG, and WebP. Choose from 3 quality levels, 3 aspect ratios, and 1-4 output images per generation. Supports reference images for style guidance. Create photorealistic images, digital art, and illustrations from simple text prompts.