Generating a landing page with GPT-3 and Dall-E2
A landing page is mainly made up some taglines and good images which capture the idea you want to convey.
And the main part of the landing page is above the fold.
So I wanted to see if I could use our AI overlords to help me create a landing page.
The goal was to provide just a prompt like “SaaS website for Yoga” or “The most luxurious car” etc and we use GPT-3 to generate the taglines needed and then use Dall-E2 to generate the related the images. Finally we can use some ready made templates to fit the text and images and get a good webpage.
Below are the results of my experiment:
- First I created an Open AI prompt to generate taglines given a product description. I got the following output.
Prompt: AI saas API
GPT3 output: {tagline : “New age API for New age SaaS and New age AI”,tagline_mini : “Don’t waste another minute, concentrate on your sales!”}
Prompt: SaaS website for Yoga.
GPT3 Output: {tagline : “Be fit and fine”,tagline_mini : “Sign up to get the best Yoga exercises customized for your body.”};
Once I was satisfied, I gave it the prompt for our web site:
Prompt: Shoes with best comfort.
GPT3 Output: {tagline : “Feet will love you for this”,tagline_mini : “You will get the most comfortable shoes as per your requirement.”};
Initially I tried to directly give the tagline as input to Dall-E2. It gave some good results, but the product was missing. For example, for “Feet will love you for this” gave the below images
So, can we improve on this?
I tried to create another GPT3 prompt which will take the tag line and change the sentence to some more descriptive for a photo.
So if someone gives a prompt like “The sky is the limit”, GPT3 will give the output as “concrete skyscraper under blue sky during daytime”.
So what does GPT3 give for our prompt “Feet will love you for this”?
It gives: woman wearing socks and shoes.
Nice. Lets give that prompt to Dall-E2. We get the following images:
I think these are much better to be used as stock photos on a landing page.
What if we add “stock photography” to the prompt. We get the following images.
Really nice.
Now we can use one of the photos above and the taglines created to create our web page.
Lots of improvements can be made. But not bad for a first attempt for AI content, right :)
Note: Dall-E2 restricts use of images in a commercial setting. So this is just for experimentation. You cannot deploy this in production right now.
