Screenshot-to-code

Why we included this project

This is one of the most-cited open-source demonstrations of a neural network generating HTML and CSS from a design mock-up. The code is organized in three iterations, starting with a Hello World proof of concept and ending with a Bootstrap version that can generalize to new screenshots using 16 domain-specific tokens. It follows the pix2code approach, and the Jupyter notebooks make the training and inference steps easy to follow. The author is upfront about the limits: only the Bootstrap variant generalizes reliably, and the training data is small and homogeneous, so treat it as a reference implementation and learning resource rather than a drop-in production tool. If you are weighing whether image-to-markup models are worth building on, this is a good place to start, with trained models available for testing.