Dataset · HuggingFaceM4
WebSight: screenshots and website HTML/CSS
HuggingFaceM4/WebSight
A large synthetic dataset of screenshot-to-HTML/CSS-code pairs for image-based UI generation.
What's inside
WebSight is a large synthetic dataset with HTML/CSS code of artificially generated English websites and corresponding screenshots. The size is 1 to 10 million examples. In version v0.1, sites are laid out in HTML + CSS without real images, in v0.2 in HTML + Tailwind CSS. The data is intended for screenshot-to-UI-code generation tasks.
What you can do with it
- Train a screenshot-to-code model
- Fine-tune VLM for HTML/CSS generation
- Evaluate UI generation from screenshots
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 18.9K
- Likes
- 412
Models trained on it
- HuggingFaceM4/Idefics3-8B-Llama3Image text → text 37.6K
- mradermacher/codium-bd-coder-v3-GGUF 1.1K
- ganjakha/codium-bd-coder-v3Text generation 690
- mradermacher/Qwen2.5-VL-3B-Instruct-Img2Code-GGUF 612
- mradermacher/Qwen2.5-VL-3B-Instruct-Img2Code-i1-GGUF 603
- PoliteChainsaw/Lewi-Surge-Preview-3B-BaseText generation 175
- HuggingFaceM4/VLM_WebSight_finetunedText generation 152
- PoliteChainsaw/Lewi-Surge-Preview-3B-Chat-EarlyText generation 93
How to load it
from datasets import load_dataset
ds = load_dataset("HuggingFaceM4/WebSight")