InfogrAIphify: Create Infographics from Articles

This is a submission for the Cloudflare AI Challenge.

What I Built

InfogrAIphify is a web app that crawls any given URL, searches for or tags, and uses their content to create a meaningful infographic for you! With it, summarization reaches new levels! ( an image :P )

The main problem with the original Python script is that it demands an openAI API key, which is paid, so when the opportunity to use a free service, well, I embraced it as an alternative.

Layout and code is a Next.js project bootstrapped with c3.

Features:

• Analyze website content and extract key points.
• Generate a visually appealing infographic.
• User-friendly interface for easy input and interaction.

Demo

The demo can be seen and here

IMPORTANT: check notes below

My Code

  cd infogrAIphify
  npm install

NEXT_PUBLIC_CF_WORKER=<WORKER URL YOU CREATED>

Journey

This time, I decided to use a CloudFlare Page starter instead and if the whole app wasn't complex enough, it would be a breeze to finish and deploy.

Some obstacles:

• The original script uses GPT4 to analyze the , which is fine, as he accepts a large input. The free LLMs of CloudFlare have a smaller limit, so I had to summarize the article using Bart first;
• BUT: Bart on CloudFlare doesn't have (at least I wasn't able to make it work this way) a setting to make the summarization larger and more complete. So the LLM would have just a small phrase to use as a basis for the infographics;
• Another problem: the Models for text-to-image available on CloudFlare aren't exactly customized to create images like those described by the LLM, something that DALL3 does flawlessly. An infographic is created, but its final result isn't as cool as if you try the Python script.
• FINALLY: it wasn't exactly a good idea to upload TWO high-usage apps on the same free-tier account, so, probably, the example may not work. Try later.

I'm sure that with more time to code, test, and check the models for better prompts, I can improve the infographic's final result. For now, as a PoC, the status of the demo is:

Multiple Models and/or Triple Task Types

Models in use:
bart-large-cnn

mistral-7b-instruct-v0.1

stable-diffusion-xl-lightning