How to identify AI art?
The Challenge
To better inform our viewers by giving them video-based content to help explain today’s complex topics, in this case, AI generated art. To create clear, engaging and visually stunning assets that support online articles. These are not just little videos, these are high quality, beautiful visual stories.
Article source: https://www.wired.com/story/how-to-spot-generative-ai-art-according-to-artists/
Research
Wired is an American magazine company that founded in San Francisco in 1993.
They publish both print and online editions, focusing on technology, culture, science and economics. They assist readers in understanding the latest technological trends and their impact on the future.
Script
How to identify AI art?
Throughout history, humans have devised tools to help create art. Recently, a powerful tool has emerged: a text-to-image generator powered by AI.
Then how do we spot AI-generated art? For now, we can pay attention to the details such as misplaced finger details, textures that are too smooth, or cross-reference what you're seeing with other credible sources.
While not at the stage of completely replacing human creativity, it would be unwise to underestimate the determination of its developers to overcome these obstacles.
Identifying AI-generated images will be challenging, but AI recognition systems and global regulations will evolve and adapt.
The value of art is what we give it. If you have a brilliant idea, AI may be the means to bring your ideas to life!
Design Approaches
Direction 1: Collage
Direction 2: Vector 2D
Direction 3: Felt Stop Motion
Style Frames
Process
Everything is done in After Effect. To make the animation look more like a stop motion, I added wiggles and constantly shifting textures to the moving elements. Below are examples of some before and after.
Like doing stop-motion animation in reality, when you place something you have a hard time not touching other elements in the frame. As the mouse passes over these tools, their placement is slightly shifted.
In real life, when the mouse clicks on a button, that button also has an interactive animation with the mouse. So I added some variations in size and position to these buttons.
Again, in stop motion, there will be no perfect animation. So I added some position shifting to the loading icon as well, to make them look more like they were placed by hand, frame by frame.
VO: For now, we can pay attention to the details such as misplaced finger details, textures that are too smooth, or cross-reference what you're seeing with other credible sources.
In previous version, I showed “misplaced finger details, textures that are too smooth, or cross-reference what you're seeing with other credible sources” all together, which could cause viewers to miss important content by not knowing where to look.
In order to everything super clear, I animated everything that’s in the voice over so that audience won’t feel lost. “Misplaced finger details” and “smooth textures” are easy to identify. However, when it comes to “cross-reference what you’re seeing with other credible sources,” I need to put the credible sources(the book in this case) next to the ai-generated picture and have them compared.
VO: Identifying AI-generated images will be challenging, but AI recognition systems and global regulations will evolve and adapt.
To have a better composition, I moved “AI/NOT AI” to the top. And have the animation showing based on voice over.
In previous version, I have METADATA and LAW to represent AI recognition systems and global regulations. In a conversation with my creative director we talked about how METADATA now also has the potential to be modified. To make sure the information is accurate, I replaced METADATA with AI SCANNER.
