From the course: DALL-E: the Creative Process and the Art of Prompting
Pushing your aesthetic - DALL-E Tutorial
From the course: DALL-E: the Creative Process and the Art of Prompting
Pushing your aesthetic
- In the last video, I gave you my ideal structure for a good prompt, but it's these two elements, and primarily the last one, that will give you the most dramatic visual differences and is the key to giving you the most consistency between your generations. So let's dive in. Visual styles. First, there are dozens and dozens of visual styles. Some are based on artistic mediums such as oil painting or photography, others are based on visual genres like anime or pop art. And styles can be regional, like Latin or Asian art, or historical periods of time, like the Victorian era. So let's demonstrate this. I'm going to start with a prompt, "A vintage toy race car on a table," and all we're going to do is change the style between several artistic mediums. We'll start with a charcoal drawing. All right, let's look at an oil painting. Pen and ink, photograph, and a wood sculpture. As you can see, it's the same subject, but with wildly different visual styles. All right, so let's take this one step further. Rather than just an oil painting, let's try subgenres like cubism, or Dutch masters, or expressionism. And once again, DALL·E produced incredibly different results, you know, capturing the nuances between each of these genres, maybe not to an art historian's liking, but still fairly well. Now let's try the same approach with photography, and explore a few subgenres here, starting with Polaroid, or pinhole photography, or lastly, and most commonly, 35-millimeter photography on film. And I absolutely love the result here, because it captures the soft focus. So as you can see, there's a wide array of styles which DALL·E knows, and knows quite well. And if you'd like, pause the screen here for a moment and take note of any that catch your eye. And remember, this list is only just a sampling of the styles that DALL·E knows, and each can be combined in fascinating ways. Alright, now let's talk about visual details. In addition to the style, there are other visual ways to customize and hone your generations. And per my prompt guide, I like to drop them in between the subject and the style. So what kind of details am I talking about? Well, here's a quick list. I'm going to start with lighting. I usually like to define the quality of light for an image, like daylight or golden hour or flash photography. Now, camera refers to the location of the camera, or how you are viewing your subject, say closeup or wide angle, or looking up from below. Color describes the tonal quality of the image, say black and white to muted colors or cool tones. And lastly, there's mood which you might use to give a more emotional description to your image. Now, it's not required to include details from each of these columns, but if you want more precision in your generations, I'd say give it a detail from one or more. To demonstrate, I've taken the prompt, "A wicker basket of fruit on a table," with the style being a photograph. And all I've changed are the details from daylight to dusk, to sepia tone, to a fisheye lens with high contrast light, and lastly, a dystopian moody image. And when you look at these, adjusting the details is like having Photoshop as an AI. I mean it's the same subject, but wildly different visual results. Now I suggest to anyone starting with DALL·E, experiment with these styles and details however you'd like. Push it. See if you can find a style that doesn't know. And if you do find one, you know, shoot me an email, I'd love to know. Alright, now let's talk about visual consistency. You know, how do you get more consistent results from one generation to the next? And the answer is, when doing a series, try to keep the details and style as consistent as possible, changing only the primary subject. For example, let's try this as my details and style. Artificial light, cool muted colors, soft-focus 35-millimeter photography. And I'll start with the subject, "A sad vintage robot in a factory." And here we go. Notice the color temperature and the mood and the visual style. All right, now let's change that subject to a full-body view of a vintage robot running through a factory. And compare the result. Pretty similar, right? Let's try two more prompts. "A close-up rear view of a vintage robot in a factory." All right, that's good. And an assembly line in a robot factory. Alright, now let's compare them all together. And what you'll see are four images that look similar enough that it feels like the same world. Alight, let's try one more, as if we're building an architectural mood board. I'm going to set the details and style to a modern architectural home built of wood and glass set in the winter mountains, morning light, and high-end architectural photography. And I'll start with the subject, "Looking outside at an expansive patio and infinity pool." Generate, and look at that. Two beautiful mood board renders. Now let's change the subject to an exterior view of an expansive patio and infinity pool. And boom, look, two more generations that are great, and all four work incredibly well together. It actually is a pretty compelling mood board. Now, in closing, it's important to remember that DALL·E treats each generation as unique. It will not remember an object from one generation to the next. Now, that might be fixed in time, but at the moment, generations are all individual images. However, with focused details, combined with a good style description, you can produce more visually synchronous results. But what happens when that's still not working and a prompt isn't working out for you at all? Well, we'll dive into that in our next video.