For the past month or so I have been enjoying using Midjourney, a platform that creates images following user-supplied directions. You reach Midjourney through Discord, and so I used my account to add the server. It wasn’t easy at first, so I did the thing that pretty much everyone does, which is to copy what other users were writing. The questions you give to an AI are called ‘prompts,’ and it is the request that the user makes to the AI to build an image or have a text answer. Gradually I began to realize that building a prompt is not really that simple, because it depends on how you write the prompts first of all, then you have to add other information, such as the background you want in the image, whether it should be realistic or cartoon-like, or indicate the type of lights to be used.
Create a realistic image of a woman sitting on a wineyard table in autumn drinking a glass of a red wine. Use a Hasselblad camera with a 85mm lens at F 1. 2 aperture setting to blur the background and isolate the subject. On the table should be a bottle of wine and a plate of chees.The wineyard should have colorful autumn leaves and vines in the background, with soft sunlight falling on the subject’s face. Use a warm and inviting lighting effect to create a cozy and inviting image. Use the Midjourney v5 with photorealism mode turned on to capture the woman’s natural beauty and grace. (Prompt of the post’s image)
You can prompt down if the image has to be a drawing and indicate the style, such as ‘Homer Simpson type’ or ‘Pixar style’. And if we want an image like a photograph we can indicate the type of camera to be used, a Sony, a Nikkon, the type of lens, the ISO. In short, the more details you provide, the better the result will be.
For me the interesting thing is trying to figure out the way to construct the question, that is, how to construct the prompt. In fact how to program it, it already talks about Prompt Engineering, the craft of constructing the prompt according to the result you want to achieve. The Prompt Engineer works also on text-based AIs, such as chatGPT or Bard, the Google’s AI.