Automator 4.11: Generate WordPress images with OpenAI

Okay, it’s only been 2 weeks since our last major release, but some exciting developments over the last few days inspired us to get a new update out. Uncanny Automator 4.11 is a smaller one for us, but it has something very cool: support for generating images with OpenAI‘s DALL-E. If you’re not familiar with DALL-E, it’s AI system that can create realistic images and art based on a natural language prompt.

Introducing DALL-E support

In the Automator 4.10 release 2 weeks ago we added OpenAI support to generate new text content from a prompt. With it, you could easily publish AI-generated blog posts, send email summaries, reply to user questions with comments (or even via an Automator chatbot) and much, much more.

Missing from all this artificial intelligence though was support for images–but not just generic image support, we wanted to make it easy to generate featured images and bring media directly into your WordPress media library. (Did we mention this is all in our free release of Uncanny Automator?)

Check out this example for a site with a listing of recipes (where recipes are custom post types):

What’s happening in the recipe example is that a user submits a form with 2 fields: a text prompt and an image prompt. The text prompt tells OpenAI to generate a recipe for that dish. The image prompt tells OpenAI to generate an image of the food item that will be used as the featured image for the post. And then the 3rd action in the recipe creates a new post that includes the AI-generated recipe and its associated image. It’s entirely automatic.

The new OpenAI action to generate an image simply requires 2 things: a prompt and an image size.

Any actions that follow this action will have action tokens available to use data from the image generation. Need the media library ID, URL or even the OpenAI image URL? No problem, they’re all available as tokens.

Here’s how we use one of those action tokens to set the featured image of our new post to use the AI image we created:

And, unlike some OpenAI image generation options out there, we take care of adding it to the media library–there’s no hotlinking involved.

Uncanny Automator 4.11 also includes some changes to better handle empty OpenAI responses. We saw from a few customers that if OpenAI couldn’t understand or reply to a prompt, it would return a blank value but flag it as successful. There is now messaging for users so it’s more clear if there was an error or OpenAI did intentionally return a blank response. If that happens, review your prompt and perhaps test it on ChatGPT so you can adjust it accordingly.

And one more thing…

The “Send an email” action can now use an optional reply-to address to direct emails replies to a different address than the sender.

We hope you find the new additions useful!