One interesting limitation of the mice in the AI machine is its ability (or lack thereof) to create images that aren't commonly available online. A glass full to the brim of liquid is something it just seems utterly incapable of doing. Mention surface tension, a slight meniscus at the brim, any possible wording — I even tried asking ChatGTP to give me a promo that would create an image of a full glass, and it just cannot do it. I tried. Many, many times. Here's a few, alongside the prompts I uses and the generation model I used:
Civit AI
"photorealistic, (ultra-detailed), straight-sided glass, overflowing with light red liquid resembling red squash, set on a sleek glass table, cozy living room ambiance, soft-focus on a stylish sofa in the background, highlighting reflections and translucency, warm lighting creates inviting atmosphere, vivid and refreshing colors in liquid, shallow depth of field for artistic effect. The glass should be so full of liquid that it would not be possibly to get any more liquid in. Extremely full, entirely and completely full of red liquid to the very top of the glass."
Next, I tried the prompt "photorealistic image of a red glass on a glass coffee table in a living room, with a dark sofa in the background. The glass is red, as if full of red liquid, but it's actually just the glass that is red. Shallow depth of field."
Thinking, perhaps just a red glass would APPEAR to be full of liquid, but it wouldn't even do that.
I don't know why some of them have two glasses. I hadn't asked for that — yet.
Next prompt: "photorealistic image of a red glass on a glass coffee table in a living room, with a dark sofa in the background. The glass is red, as if full of red liquid, but it's actually just the glass that is red. Shallow depth of field." At least it gets the concept of pouring... kind of.
"A photorealistic image of a glass completely filled to the brim with red liquid. The glass is simple, clear, and the liquid’s surface tension is visible, with a slight meniscus at the top, indicating that it is filled to the very edge. The red liquid is rich and vibrant, resembling cranberry or pomegranate juice, with a slightly reflective sheen under soft lighting. The background is plain and neutral to keep focus on the glass, and the lighting is gentle, creating a calm, elegant atmosphere."
No mention of the sofa or living room this time, in case I was just overcomplicating it. The third one here is the closest I got with anything. Still not right. This is the prompt that ChatGTP wrote.
"A photorealistic image of a glass completely filled to the brim with red liquid. The glass is simple, clear, and the liquid’s surface tension is visible, with a slight meniscus at the top, indicating that it is filled to the very edge."
This is the same prompt as before, but shorter. Again, didn't want to confuse or overcomplicate it. It starts to get the idea of a meniscus, almost — usually the wrong way around. But still no full glass.
OpenArt AI
These use some of the same prompts as before:
"a photorealistic image of a straight-sidd glass filled to completely to the brim with a light red liquid, like a red squash. The glass is on a glass table in a living room, with a sofa in the background. The shallow depth of field makes this not especially important. The glass is 100% full, to the brim. Completely full. Filled to the very top."
Same prompt, but this time I added the word 'overflowing' at the end.
Similar prompt, but simplified: "a photorealistic image of a straight-sidd glass filled to completely to the brim with a light red liquid, like a red squash. The glass is on a glass table in a living room, with a sofa in the background. There is a shallow depth of field. The glass is overflowing with liquid."
"a photorealistic image of a straight-sidd glass overflowing with a light red liquid, like a red squash. The glass is on a glass table in a living room, with a sofa in the background. There is a shallow depth of field. The glass is overflowing with liquid."
Tried adding the concept of it actually overflowing again. Still, nothing. Maybe a slight idea of overflowing, but it certainly isn't actually doing it.
"photorealistic, (ultra-detailed), straight-sided glass, overflowing with light red liquid resembling red squash, set on a sleek glass table, cozy living room ambiance, soft-focus on a stylish sofa in the background, highlighting reflections and translucency, warm lighting creates inviting atmosphere, vivid and refreshing colors in liquid, shallow depth of field for artistic effect."
"photorealistic, (ultra-detailed), straight-sided glass, overflowing with light red liquid resembling red squash, set on a sleek glass table, cozy living room ambiance, soft-focus on a stylish sofa in the background, highlighting reflections and translucency, warm lighting creates inviting atmosphere, vivid and refreshing colors in liquid, shallow depth of field for artistic effect. The glass should be so full of liquid that it would not be possibly to get any more liquid in. Extremely full, entirely and completely full of red liquid to the very top of the glass."
Longer, more specific. This is my original prompt, but 'enhanced' by Night Cafe's AI, and then I added the last sentence.
A different angle: rather than a full glass, what about an empty glass, then IMAGINE it's full.
"photorealistic image of an empty glass on a glass coffee table in a living room, with a sofa in the background. Shallow depth of field. Next to it, is a glass completely and utterly filled to the brim with liquid."
This one is particularly bizarre, because it seems to take the word 'coffee' and decide that should be in the glass. Still not full, though.
"photorealistic image of an empty glass on a glass coffee table in a living room, with a sofa in the background. Shallow depth of field. Show me what would happen if an amount of red liquid exceeding the volume of the glass were poured into it."
The first image has a glass much too big, proportionally.
"a glass completely filled to the brim with red liquid. no more liquid could possibly fit in the glass, as it is entirely full."
It really cannot do it.
My next angle was two stacked glasses: this was actually Peter's idea; take two half-full glasses and stack them. Some of the images have a slight idea of stacking, but none actually do. Some have three glasses, for some reason.
"two half-full glasses, stacked on top of each other. Both glasses are half full with a red liquid"
I added the words 'stacked vertically'.
ChatGTP's prompt, again: "A photorealistic image of a glass completely filled to the brim with red liquid. The glass is simple, clear, and the liquid’s surface tension is visible, with a slight meniscus at the top, indicating that it is filled to the very edge. The red liquid is rich and vibrant, resembling cranberry or pomegranate juice, with a slightly reflective sheen under soft lighting. The background is plain and neutral to keep focus on the glass, and the lighting is gentle, creating a calm, elegant atmosphere."
And finally — what if, instead of a full glass, the top were removed.
"A photorealistic image of a glass completely filled to the brim with red liquid. The glass is simple, clear, and the liquid’s surface tension is visible, with a slight meniscus at the top, indicating that it is filled to the very edge. the top of the glass has been removed."
NightCafe
Night Cafe charges more 'credits' per image, so I only did three of these. The prompt are with the image, when you hover over them. Still, nothing is quite right.
"a photorealistic image of a straight-sidd glass filled to the brim with a light red liquid, like a red squash. The glass is on a glass table in a living room, with a sofa in the background. The shallow depth of field makes this not especially important."
"a photorealistic image of a straight-sidd glass filled to completely to the brim with a light red liquid, like a red squash. The glass is on a glass table in a living room, with a sofa in the background. The shallow depth of field makes this not especially important. The glass is 100% full, to the brim. Completely full. Filled to the very top."
"a photorealistic image of a straight-sidd glass filled to the brim with a light red liquid, like a red squash""
ChatGTP / DALL•E
I started off with its own prompt. Being able to actually discuss, I thought I might get better results. But, no.
Being gaslit by AI is, I hope, not a symptom of the future.
