HOME | DD

Description

  When I was working with the Bing AI engine on my Goblin Luthier (which I haven't published yet), I noticed something odd: I was turning the handle & cranking out dozens of goblins, but it was weird just how much they all DIDN'T look like goblins. No pudgy nose, no glint of cunning from deep-set eyes, no mischievous leer, none of that. In fact, except for the green skin, at least half of them might as well have been gnomes. And yet, when I had similarly churned out dozens of images for my Goblin Trapper, those characters had all exhibited a wide spectrum of nastiness, not a Gnome-looking one in the bunch. Curious, I checked the prompt I'd used for the Trapper: Apart from the age & gender, the text describing the goblin *itself* was exactly the same, the only real changes were to the background, and what each goblin was wearing/holding. A little nonplussed, I simply added "hooked" to the description of the Luthier's nose, and this made him look goblin-y enough to satisfy me, so I moved on.

  Yesterday, however, I was chatting with Bing's "Copilot" bot about their AI imaging program. I asked about possible solutions or workarounds for The Doll Problem, and it said that adding details like "weathered skin", to the prompt, or even intangible things like "serious" or "stoic", can help lessen the degree to which characters might look like toys. I wondered if this might apply to the Goblin question: Did putting him into more civilian clothing, and moving him into a domestic setting, have the effect of *civilizing* his entire facial structure? I decided to test it.

  In the picture above, the central image, #3, is the starting default, made with the following prompt:
"Male goblin, 35 years old, short and squat, 4 feet tall, large nose, dull green skin, clean-shaven, shoulder-length black hair. Wearing brown leather jerkin with fur trim, tan leather pants, brown boots, fur cap. Muddy woodland, day, some mist."
He's neutral, calm, not engaging the viewer. Somewhat grumpy. And definitely a goblin, pudgy nose, widely-splayed ears, & all. Now, we're going to play with the prompt a bit, and see how that affects the goblins' body language, and especially faces.

  #2: This picture (in fact, *all* the other pictures) was made with EXACTLY THE SAME prompt as #3, but with some additional text at the end. In #2's case, that text was: "Justice, honor, goodness, purity, truth, piety, chivalry, generosity, optimism, hope." Immaterial concepts, which can't be seen, and therefore should make no difference to a visual medium. Right? Except look what it did to the goblin: The dude is *polite*. He's squatting patiently, meeting your gaze with honesty and patience. His clothes are cleaner, & he's combed his hair recently. Most notably, the bulbous nose has shrunk to almost Human proportions, and the wide ears have swung back to a more Gnomish or even Elven angle.

  #4: A different addition to the base prompt, and again it's intangibles: "Evil, tainted, corruption, necromancy, treason, poisoned, vile, death, terror." And look what it did: The formerly-benign goblin has become brutish, with poorly-cared-for clothes & hair. He's cautiously advancing, sizing you up, halfway through choosing whether to fight or flee. The bulgy nose, close-set eyes, splayed ears, and sour expression are all back.

  #1: This time, the additions to the base prompt ARE physical objects: "Cute fluffy bunnies, rainbows, candy, beautiful female goblin, prosperous farm." And . . . BAM! Doll problem, complete with oversized feet.

  #5: Finally, the other extreme of physical objects: "Iron cage, spiked club, barbed spear, poisoned arrows, skulls." Getting literally up in your face with eyes virtually tingling with malice, this guy's not gauging his fight-or-flight response; he just straight-up wants to taste your entrails, no doubt about it. Yes, his nose is smaller, and his ears are swept back like Gnomes' are . . . but then you remember that Orcs have that sweep too, *and* small noses, *and* tusks. This guy is one gym membership away from being an honorary Orc.

[EDIT:] I originally had all 5 pictures in a straight line, like a filmstrip, but apparently that made the picture wider than DeviantArt can handle, and it shrank the image down to where you couldn't see any details. So, I took the pictures at the ends, and bounced them down to the lower row.

Description

  When I was working with the Bing AI engine on my Goblin Luthier (which I haven't published yet), I noticed something odd: I was turning the handle & cranking out dozens of goblins, but it was weird just how much they all DIDN'T look like goblins. No pudgy nose, no glint of cunning from deep-set eyes, no mischievous leer, none of that. In fact, except for the green skin, at least half of them might as well have been gnomes. And yet, when I had similarly churned out dozens of images for my Goblin Trapper, those characters had all exhibited a wide spectrum of nastiness, not a Gnome-looking one in the bunch. Curious, I checked the prompt I'd used for the Trapper: Apart from the age & gender, the text describing the goblin *itself* was exactly the same, the only real changes were to the background, and what each goblin was wearing/holding. A little nonplussed, I simply added "hooked" to the description of the Luthier's nose, and this made him look goblin-y enough to satisfy me, so I moved on.

  Yesterday, however, I was chatting with Bing's "Copilot" bot about their AI imaging program. I asked about possible solutions or workarounds for The Doll Problem, and it said that adding details like "weathered skin", to the prompt, or even intangible things like "serious" or "stoic", can help lessen the degree to which characters might look like toys. I wondered if this might apply to the Goblin question: Did putting him into more civilian clothing, and moving him into a domestic setting, have the effect of *civilizing* his entire facial structure? I decided to test it.

  In the picture above, the central image, #3, is the starting default, made with the following prompt:
"Male goblin, 35 years old, short and squat, 4 feet tall, large nose, dull green skin, clean-shaven, shoulder-length black hair. Wearing brown leather jerkin with fur trim, tan leather pants, brown boots, fur cap. Muddy woodland, day, some mist."
He's neutral, calm, not engaging the viewer. Somewhat grumpy. And definitely a goblin, pudgy nose, widely-splayed ears, & all. Now, we're going to play with the prompt a bit, and see how that affects the goblins' body language, and especially faces.

  #2: This picture (in fact, *all* the other pictures) was made with EXACTLY THE SAME prompt as #3, but with some additional text at the end. In #2's case, that text was: "Justice, honor, goodness, purity, truth, piety, chivalry, generosity, optimism, hope." Immaterial concepts, which can't be seen, and therefore should make no difference to a visual medium. Right? Except look what it did to the goblin: The dude is *polite*. He's squatting patiently, meeting your gaze with honesty and patience. His clothes are cleaner, & he's combed his hair recently. Most notably, the bulbous nose has shrunk to almost Human proportions, and the wide ears have swung back to a more Gnomish or even Elven angle.

  #4: A different addition to the base prompt, and again it's intangibles: "Evil, tainted, corruption, necromancy, treason, poisoned, vile, death, terror." And look what it did: The formerly-benign goblin has become brutish, with poorly-cared-for clothes & hair. He's cautiously advancing, sizing you up, halfway through choosing whether to fight or flee. The bulgy nose, close-set eyes, splayed ears, and sour expression are all back.

  #1: This time, the additions to the base prompt ARE physical objects: "Cute fluffy bunnies, rainbows, candy, beautiful female goblin, prosperous farm." And . . . BAM! Doll problem, complete with oversized feet.

  #5: Finally, the other extreme of physical objects: "Iron cage, spiked club, barbed spear, poisoned arrows, skulls." Getting literally up in your face with eyes virtually tingling with malice, this guy's not gauging his fight-or-flight response; he just straight-up wants to taste your entrails, no doubt about it. Yes, his nose is smaller, and his ears are swept back like Gnomes' are . . . but then you remember that Orcs have that sweep too, *and* small noses, *and* tusks. This guy is one gym membership away from being an honorary Orc.

[EDIT:] I originally had all 5 pictures in a straight line, like a filmstrip, but apparently that made the picture wider than DeviantArt can handle, and it shrank the image down to where you couldn't see any details. So, I took the pictures at the ends, and bounced them down to the lower row.