Close Menu
Versa AI hub
  • AI Ethics
  • AI Legislation
  • Business
  • Cybersecurity
  • Media and Entertainment
  • Content Creation
  • Art Generation
  • Research
  • Tools
  • Resources

Subscribe to Updates

Subscribe to our newsletter and stay updated with the latest news and exclusive offers.

What's Hot

One year since “Deep Seek Moment”

March 5, 2026

The most cost-effective AI model ever

March 4, 2026

Google’s industrial robot AI Play makes physical AI a priority

March 4, 2026
Facebook X (Twitter) Instagram
Versa AI hubVersa AI hub
Thursday, March 5
Facebook X (Twitter) Instagram
Login
  • AI Ethics
  • AI Legislation
  • Business
  • Cybersecurity
  • Media and Entertainment
  • Content Creation
  • Art Generation
  • Research
  • Tools
  • Resources
Versa AI hub
Home»Tools»Unlocking conversion of web screenshots to HTML code using WebSight dataset
Tools

Unlocking conversion of web screenshots to HTML code using WebSight dataset

versatileaiBy versatileaiJuly 1, 2025No Comments3 Mins Read
Share Facebook Twitter Pinterest LinkedIn Tumblr Reddit Telegram Email
#image_title
Share
Facebook Twitter LinkedIn Pinterest Email

In the world of web development, turning a design into a functional website usually involves a lot of coding and careful testing. What if we simplified this process and made it easier and faster to convert web designs to work websites? WebSight is a new dataset intended to build AI systems that can convert screenshots into HTML code.

Challenge

Turning website designs or screenshots into HTML code usually requires an experienced developer. But what if this could be more efficient? Motivated by this question, we investigated how to use vision language models (VLMs) in web development to create low-coded solutions that improve efficiency.

Today, the main challenge to that goal is the lack of high quality datasets tailored for this task. WebSight aims to fill that gap.

WebSight: Large synthetic dataset of screenshot/HTML code pairs

In January 2024, we introduced WebSight-V0.1, a synthetic dataset consisting of 823,000 pairs of HTML code and corresponding screenshots. This dataset is designed to train AI models to process and transform visual web design into functional HTML code. By focusing on synthetic data, we bypassed the noise and complexity that are common in real HTML, allowing AI models to learn efficiently.

In addition to community feedback, we updated our dataset to WebSight-V0.2 following our initial release and construction, and introduced significant improvements. These extensions feature switching to Tailwind CSS instead of traditional CSS, using real images in screenshots. Additionally, we have expanded our dataset to 2 million examples.

Examples of web pages included in WebSight.

Sightseer: A fine-tuned model with WebSight

Using the WebSight dataset, we’ve fine-tuned future Foundation Vision-Language models to get Sightseer, a model that can convert web page screenshots into functional HTML code. Stightser further demonstrates the ability to incorporate images into generated HTML that are very similar to those in the original screenshot.

Comparing the original web page (input) on the left, and rendering of code generated by Sightseer (output) on the right.
Comparing the original web page (input) on the left, and rendering of code generated by Sightseer (output) on the right.

Towards more powerful tools unlocked by visual language models

By repeating WebSight, our goal is to build a more capable AI system that simplifies the process of turning UI design into functional code. This reduces developer iteration times by quickly converting paper UI sketches into functional code, making this process more accessible to non-developers. This is one of many real applications of visual language models. With Open-Source Websight, the community encourages us to work with us to build more powerful tools for UI development.

resource

author avatar
versatileai
See Full Bio
Share. Facebook Twitter Pinterest LinkedIn Tumblr Email
Previous ArticleAI Art Protection Tools put creators at risk
Next Article Silicon Valley Insider is revealing AI companies like cults
versatileai

Related Posts

Tools

One year since “Deep Seek Moment”

March 5, 2026
Tools

The most cost-effective AI model ever

March 4, 2026
Tools

Google’s industrial robot AI Play makes physical AI a priority

March 4, 2026
Add A Comment

Comments are closed.

Top Posts

Open Source DeepResearch – Unlocking Search Agents

February 7, 20259 Views

Improving the accuracy of multimodal search and visual document retrieval using the Llama Nemotron RAG model

January 7, 20267 Views

Google’s industrial robot AI Play makes physical AI a priority

March 4, 20264 Views
Stay In Touch
  • YouTube
  • TikTok
  • Twitter
  • Instagram
  • Threads
Latest Reviews

Subscribe to Updates

Subscribe to our newsletter and stay updated with the latest news and exclusive offers.

Most Popular

Open Source DeepResearch – Unlocking Search Agents

February 7, 20259 Views

Improving the accuracy of multimodal search and visual document retrieval using the Llama Nemotron RAG model

January 7, 20267 Views

Google’s industrial robot AI Play makes physical AI a priority

March 4, 20264 Views
Don't Miss

One year since “Deep Seek Moment”

March 5, 2026

The most cost-effective AI model ever

March 4, 2026

Google’s industrial robot AI Play makes physical AI a priority

March 4, 2026
Service Area
X (Twitter) Instagram YouTube TikTok Threads RSS
  • About Us
  • Contact Us
  • Privacy Policy
  • Terms and Conditions
  • Disclaimer
© 2026 Versa AI Hub. All Rights Reserved.

Type above and press Enter to search. Press Esc to cancel.

Sign In or Register

Welcome Back!

Login to your account below.

Lost password?