abidlabs/vlm-screenshot-to-webpage
Screenshot → webpage, two VLMs side by side
A gr.Workflow that sends one webpage screenshot to two recent vision-language models and compares what they generate. Each branch returns the model's HTML and a screenshot of that HTML rendered in a real browser, so the two can be compared visually against the original rather than by reading their code.
Both run as plain model nodes through HF Inference Providers — no custom inference code. The only bound Python function is the renderer, since driving a browser isn't an inference task.
Sign in with Hugging Face to run the workflow; inference uses your own account.
Running a build of Gradio from PR #13666, which makes image-text-to-text model nodes work by calling vision-language models through chat completions instead of the retired visual_question_answering task.