StyleForge Neural Style Transfer

Result appears here

Upload a content image and a style
reference, then click Transfer Style.

Examples

See what it produces

Each row shows the content image, the style reference, and the final output.

Content Content
Style Style
Output Result
Content Content
Style Style
Output Result
FAQ

Questions answered

No — the decoder is trained from scratch. Only the VGG encoder uses frozen ImageNet weights as a feature extractor.
This demo is currently available as a free experience with no account required.
Paintings with strong, distinct textures — impressionist, watercolour, cubist, ink sketch. Photorealistic style images produce subtle results.
Alpha blends content and style features. 1.0 = full style applied. 0.0 = original content unchanged. Values around 0.7–0.9 give the best balance.
Python, PyTorch, and Flask. The model uses VGG feature extraction with Adaptive Instance Normalisation (AdaIN).