AI Generated Photos: IT’S TOO REAL
Education
Introduction
Hey everybody, Josh here. What if I told you that half of the video you're watching right now is AI-generated? You might be asking, "What are you exactly talking about?" Let me show you.
That painting on the wall? It doesn't exist. That trophy on that shelf? That doesn't exist either. In fact, all of this space around me doesn't actually exist. What we're left with is this: the original shot that came out of my camera. This is all made possible by Adobe Generative AI.
What is Generative AI?
Generative AI is a type of artificial intelligence system capable of generating text, images, and other media in response to prompts. These AI models are trained on hundreds of thousands of samples of real-life datasets and then use those datasets to generate new creations. We've seen ChatGPT and DALL-E, but now with Adobe Generative AI, we can get some real-life use out of this technology. I’ve broken it down into three primary use cases:
1. Object Removal
This is one of the most classic use cases for Photoshop. Before AI, we had to rely on tools like the Clone stamp tool and content-aware fill. With generative AI, the Clone stamp tool is essentially dead. You can just select what you want to be deleted, click a button, and it will be gone.
For example, in a photo I took in Japan full of people, I was able to remove everybody from the scene in just a couple of minutes. Even though it's in a beta stage, it's really impressive.
2. Object Generation
This feature can be hit or miss. At times, it gives some corny-looking results, such as robots and hot air balloons looking kind of lame. However, it can also impress, like the trophy in the intro shot. The AI had to get several things right to make it look believable. I also tried generating a puddle onto the ground in a motorcycle shot. Some results were better than others, and when zoomed out, it looks pretty good. This feature will improve as the AI evolves.
3. Scene Extension
This feature blew me away. The AI blends the existing real-life pixels from your original photo into something completely AI-generated. There are many use cases, such as extending a landscape photo to fill a phone screen more effectively for Instagram or digitally extending a set.
Adobe struggles with human faces, which makes sense because human faces are one of the hardest things for AI to make look real. Often, generated faces end up looking like a blur or have something off about them, making it obvious they're not real.
Use Cases and Limitations
Recently, two AI-generated photos went viral on the internet: one of the Pope dripped out in Balenciaga and another of Trump getting arrested. These photos weren't generated with DALL-E or Adobe Generative AI. Adobe has specific user guidelines that prohibit generating offensive or impersonating content. Thus, these images were made using MidJourney, another AI generation tool that creates even more realistic and detailed images.
MidJourney: A More Advanced Tool
MidJourney generates faces impressively, getting rid of the uncanny valley effect. In the right circumstances, it convinces the world that these images are real. You can upload a photo of your face, give it a reference photo, and it can perfectly place your face onto that reference photo with a click of a button.
You can generate photos inspired by certain movies or directors, such as "Harry Potter directed by Wes Anderson" or "Star Wars directed by Wes Anderson." These images could theoretically be printed out and hung on a wall. Under current US law, copyright laws only apply to works created by humans. Hence, images created by AI are not legally owned by anyone.
The Future and Ethical Concerns
This leads to moral and legal questions like: Who do these images belong to? More importantly, how do we tell what's real and what's not? Imagine the ease with which fake evidence could be created in the legal system. AI-generated video is the next frontier, and we will soon live in a world with human art mixed in with AI-generated text, photos, videos, and music.
Right now, we are getting a glimpse into what this world is going to look like, and it's all moving very fast.