Build AI powered Sites with Divi

Our initial step is to select a foundational theme. For this purpose, we’ve chosen Divi due to its exceptional dynamic content support, simplifying the creation of an AI-powered front end. For our example, we’ve opted for the Pizzeria theme.

With Divi, incorporating dynamic content as variables is a breeze, and you can supply those values through custom meta fields. These dynamic components can subsequently be designed to align seamlessly with your existing brand aesthetic.

You can preview the template’s appearance even when there is no live content to display.

Adding static branding elements around dynamic components to align with your brand is straightforward. In this instance, we’re incorporating a static mascot onto the page.

MidJourney AI is employed to generate mascot images. We invite you to explore more of these conversational agent designs on our GitHub page. All these are distributed under the Creative Commons license, hence are available for free use.

After generating the mascot, we seamlessly integrate it into our pages, substituting the pre-existing elements from the default template.

Next, we employ the Divi Builder utilities to aptly position and fashion our images.

An added perk of the Divi Builder is the feature that lets you fashion your elements – be it images, rows, columns, or sections – in a user-friendly drag and drop interface.

Every single piece of content on this page, including the text you’re reading now, was produced by ChatGPT. There are several versions and approaches to leveraging the AI, each with its own advantages and costs. No matter which method you choose, having well-crafted prompts is essential. In our case, we make use of Google Sheets to generate and arrange the prompts for the first version.

When all the individual elements were assembled, we employed intricate concatenate formulas to develop a single, comprehensive prompt for the AI. Depending on the precision of your instructions, these prompts can become quite lengthy. If your format demands high specificity, expect the prompts to be longer and more detailed.

Subsequently, we input the prompts into ChatGPT. There are different approaches to do this: manually via the web application – an excellent choice for smaller content volumes, or through the ChatGPT API. The latter method is accessible for all users who wish to generate content using either the v4 model or the v3.5 Turbo model.

Following that, we acquire and structure the data. This can be achieved via BASH or Python when utilizing the API. Interestingly, ChatGPT has the capability to script the code required for parsing the data into your preferred layout.

All that’s needed is to collect the raw data, structure it, provide it as a sample to ChatGPT, and allow the AI to handle the strenuous work.

Subsequently, we mold and prepare the data for publication. In this step, we employ coding to duplicate our template, generating a blank page for the purpose of inputting our processed data.

Leveraging WPCLI, we transmit the formatted data to the WordPress site, ensuring all data is arranged according to this structure. You can discover more examples of WPCLI integration on our GitHub repository.

The conclusive phase involves executing the script and importing the data. Given WPCLI’s extensive capabilities, processing times may vary, contingent upon your data volume.

We initiate the process with image prompts that are also formulated by ChatGPT. In this particular instance, we utilized it to create batch image prompts. The result was a collective production of nearly 1100 images, which are readily accessible on our GitHub.

Now, MidJourney also offers an option where you can input example images, and it reverse-engineers prompts for you. This approach proves highly effective as each image you feed into it generates four prompts, each of which then creates four images. So, by providing one example image, you end up with a total of 16 images. This significantly amplifies your chances of finding a remarkable image.

You have the option to either directly generate images from the provided prompts or fine-tune them according to your specific needs.

However, be mindful of the generated output as this method might lead to unexpected “Hallucinations.”

For instance, an image of a skull might surprisingly emerge from a stock photo prompt of a hamburger and french fries. While this is an outlier, it’s crucial to remember that this approach necessitates thorough review to ensure accuracy and relevancy.

Prompts can be modified to align with your image aspect ratios. For instance, this website primarily adopts a 16:9 (Landscape) aspect ratio. In MidJourney, this is controlled with the –ar flag, enabling you to maintain a consistent visual appearance across your site.

Upon completion, it’s advisable to optimize and organize your images. We utilize Adobe Creative Cloud for this process, performing batch processing to refine and name the images.

Once optimized, the images are uploaded to GitHub, available for free usage under the Creative Commons license, providing an organized and accessible resource for your visual content.

Subsequently, we transfer the images to WordPress. There are various methods available for this, including uploading directly through the wp-admin interface, using WP-CLI from the command line, or employing one of the numerous plugins available.

Choose the approach that best suits your workflow.

Keep in mind that large quantities of images can consume substantial storage space and bulk importing can put a strain on your server. It’s advisable to carry out this process in stages to prevent any potential complications due to high resource usage.

The batch generation feature in MidJourney requires the use of “Fast Processing Time,” but it allows you to produce a vast amount of starting point images while using a limited fast processing time.

However, it’s important to note that not all MidJourney plans include the “relaxed” processing mode, which is primarily used for our bulk generation tasks.

Concluding the process, once all data is transferred to WordPress, we perform the final customization. Given that using a template can result in highly similar-looking pages, we have developed a workaround.

By generating multiple variations of each section, we ensure a diverse mix of page types in the final product, enhancing the uniqueness of each page.

Subsequently, we simply remove the sections that don’t align well with a specific page. This approach proves to be faster than using multiple templates, but it does result in some rather large pages prior to trimming the content. Therefore, it’s important to remain mindful of resource constraints associated with your hosting service.