Which Way is the Uncanny Valley?



An Internet brouhaha recently caught my eye, about a twitter thread posted by a musician named Supercomposite, who claims to have accidentally used image generating AI to generate the world's first digital cryptid.  The image is of an older woman whom Supercomposite has named Loab.  Supercomposite claims that this image is adjacent to a bunch of gory and disturbing images in the AI's understanding of the world, and that crossbreeding this image with other images produces images that are horrific, disturbing, desolate, and bleak. You can see the thread here.

Leaving the hoopla about digital cryptids aside, as an artist, I'm more interested in the application of this to horror art.

The uncanny valley is a movie term that describes the creepy effect that occurs when an animated character is rendered too realistically to be a cartoon, but not realistically enough to look real, with the result that it looks like an animated corpse.  Animated film makers try to avoid the uncanny valley, but if you're a horror artist, you may want to plot a course directly into it.

I was curious whether the creepazoid factor in the AI-generated images would survive being redrawn, or whether it was dependent on the photographic realism of the source image.  I've had this problem in the past when trying to use Artbreeder -generated images as reference.  It became evident very quickly that Artbreeder does not understand form and volume; it is collaging bits of flat images together, and the effect falls apart as soon as you redraw it and it is no longer a photograph.  If over-rendering is the key that unlocks the door to the uncanny valley, just how over-rendered does it need to be?  Does it need to look photographic, or is a pencil drawing good enough?

You can judge for yourself, but I think the horror survived mostly intact.

Comments

Post a Comment

Popular posts from this blog

Image
  Doing Fast Perspective Setups in the Field Part 5: Tips and tricks And now, a few helpful facts. To find the center of a rectangle, you can draw an x from corner to corner. This also works if the rectangle is in perspective. This is how you find the perspective center of a rectangular object. You can extend this fact to place a series of regularly spaced objects, such as telephone poles, out to the horizon, in perspective.  To freehand an ellipse, start by drawing the perspective rectangle that bounds the ellipse. Then draw an x to find the perspective center of the rectangle, and draw a plus through the center. Mark each arm of the x about a third of the way in from the edge, and now you have all the points you need to freehand the ellipse. This is a very useful technique for drawing things like car tires. Figures of equal height will intersect the horizon line at the same point on their bodies. So if the horizon line is at head level, it will intersect the heads of all the...
Read more
Image
New Project: Folk Arcana  This is going to be a system-agnostic TTRPG supplement of folk magic items, inspired by the collection in the Oxford Ashmolean Museum . These are actual artifacts that people in the UK discover inside their walls, while remodeling old houses. So here's the idea: it's not so much that magic items in D&D are overpowered (although, IMO, they are), it's that they're obsessed with power. Their conception of magic is very tactical-wargame-y; this sword gives you a combat bonus, that potion gives you some hit points back, that sort of thing. These items are going to be more like talismans you might buy from a local witch or druid; they are meant to function more as writing prompts than as power widgets in a video game. Each item has a narrative behind it, and a down-side to using it. When you use magic, you are tampering with forces beyond our ken. You can't, as they say, get the genie back in the bottle. Even minor magic should not be taken l...
Read more
Image
Doing Fast Perspective Setups in the Field Part 3: The Grid Now, finally, we are going to talk about how to freehand a perspective grid in the field. I actually don't like to start a drawing with the grid on a blank page, because it is difficult to envision how things are going to come out. Instead, I freehand the most important object in the scene, then reverse engineer the grid to fit it, so I know that the grid will make the most important object come out the way I want.  But before you do anything, start by placing the horizon line where you want the eye level to be. Don't place the horizon line too high. If the point of view is higher than the height of an average person, the picture will end up looking strange because we do not normally see the world fro this point of view. Unless you want it to look like a picture that was taken by a flying drone, place the horizon at eye level, or even lower. Placing it at waist level has a dramatic effect. Next, freehand the most impor...
Read more