People, portraits, and photorealism - DALL-E Tutorial

- [Instructor] One of the most common requests for AI is generating realistic versions of our human selves. I believe there's something intrinsically fascinating for us in witnessing a computer program attempting to capture that nuanced humanity and also watching it fail, as it creates bizarre and often comical errors. Oh man, this, this poor guy. So in this video, we're going to discuss the various techniques I've learned through experience. All right, introduction to realistic portraits. Now, my first rule of thumb for the best humans is to keep it simple. Remember, DALL-E is trained on millions and millions of reference photos. And of those, thousands and thousands are photos of people in basic settings. So using prompts that tap into those sources will yield some incredible results. Here, let's start with an example. I'm going to type in the prompt. A happy grandmother over a great background. and the aesthetic or the style will be a photograph. And look, she looks happy, I mean, she looks great. All right, let's try another example, but I'm going to add a few more details this time. A balding older salesman in an old suit, over a gray background, again with a style photograph and here we are and mean they're, they're pretty impressive. I mean, these guys look, the wrinkles look great. And if I apply the same philosophy to other simple descriptions, the results are equally good. As you can see here, with the details and clothing of these punk rockers, and here's one of my favorites, this rugged cowboy. Alright, so let's go on to more advanced techniques. What I've shown is successful generic people, but what if we want to get a little bit more specific? How does DALL-E hold up? Well, let's start with something in the fashion realm and here's where I'm going to add a lot more detail. A glamorous young woman with a mod haircut wearing round designer sunglasses and a vintage pattern dress, over a white background, and again, the style being photography, but this time we're going to add fashion. Wow, these, these all look pretty good, but they're still in the studio. So let's change the prompt slightly. I'm going to change over a white background, now into walking in New York City. You know, let's get 'em out into the real world. And here we go, they look pretty good, but now that I'm adding more details, we're also going to see a few more misses with the hits. For example, some of the hair and the expressions look great and the clothing, the looks-- Honestly, the clothing looks great and almost all of 'em, but now you're going to see some errors. Like, for example, with these two, so now she's wearing two sets of glasses for some reason and this one on the right, I mean, yes she looks like she's almost made a plastic. So realize when you add more details, there is a greater chance for these unexpected errors. And if you're credit conscious, you might want to just watch out for this. Alright, now let's get into some of my specialized techniques. Now, there are a few key words and phrases I find that are excellent additions when going for more realistic and visually interesting generations of people. They are closeup or extreme closeup, depth of field or selective focus and also I love a little soft focus. So let me show you some examples. We'll put this to the test. All right, I'm going to start with the prompt: a sad man wearing a bowling shirt who's sitting alone at a diner, again, as a photograph. Also, please don't ask where I get these prompts. I mean, I really hope this is a judgment free zone but I'm using as an example because I what I love is when you see the generation, it does capture that moment of loneliness at a diner. I think, you know, the expressions and the compositions I love, but they're still kind of wide shots or medium shots. But notice what happens when I add closeup at the start and depth of field into the details, the shot becomes much more interesting. And when you see this comparison, you can see the differences as I've now directed the camera. So let's try this again with another example. Let's type in the prompt, a wide shot of a punk rock man walking in London, photograph. And the result looks great, I mean, it's a nice full body shot of our rocker. And here's the same prompt with closeup at the start, plus depth of field added. And you can see it looks much more cinematic and almost like a documentary style shot And think about this as you experiment with your own generations. What type of look what kind of feeling would you like? You absolutely can direct the camera when generating people. Alright, so now you know some of my best tricks for portraits which means you too can make happy grandparents in ugly holiday sweaters on vacation in Paris, which I'm pretty certain is the reason DALL-E was invented. (Instructor laughs) Okay, up next, let's apply these advanced learnings in a new challenge.