The geoduck might look great, but it was pronounced wrong.
And the Cybertruck being lousy is just accuracy to the real thing.
This is fun. Found a prompt on Facebook, originally intedned for Gemini Nanobanana
Summary
Create an image of a detailed and intricate unfiltered monologue of your real thoughts on AGI in fountain pen blue ink, on it scrawl corrections in marker pen, there are doodles and weird oddities scrawled, you cut out and stick on lots and lots of photo extracts from magazines to show the point!!! sometimes you write on them too
Gemini
Copilot
Modified it to be more generic
Summary
Create an image of a detailed and intricate unfiltered monologue of your thoughts in fountain pen blue ink, on it scrawl corrections in marker pen, there are doodles and weird oddities scrawled, you cut out and stick on lots and lots of photo extracts from magazines to show the point!!! sometimes you write on them too
Gemini
Copilot
And made a couple of Sora videos
I was doodling at a bar…
And decided to ask AI to make a better version of it. It was pretty good!
I love that you can mix both a hand-drawn sketch and a text prompt, like “Can you please make a photo realistic version of this coat of arms? It’s a dog-goose hybrid, with a cowprint patterning and a Christmas Santa hat. The upper right and lower left quadrants should be striped green.”
I asked ChatGPT to describe that image in heraldry terms:
Quarterly argent and vert; in full the charge of a hybrid beast statant affronté, consisting of the body and legs of a goose proper, spotted sable, topped with the head and elongated neck of a hound proper, langued gules, the whole adorned with a Santa cap gules trimmed argent.
IANA heraldry expert, but isn’t “argent” the word for silver? ISTM the santa hat and the beast are more white than silver. Once can debate the shade of silver / gray / white of the upper left & lower right quadrants.
Huh… I guess a picture really is worth a thousand words.
I asked it to generate an image based on that description alone:
Right idea, but I guess the description lacked sufficient detail about the hound’s head or the nature of the “spotted sable” looking like cow prints.
I also had no idea what any of the heraldric text meant. Its explanation:
AI explanation of AI description of arms
This is a heraldic blazon written in formal armorial language. In plain English, it describes a shield and the creature displayed on it:
Shield (background):
- Quarterly argent and vert — The shield is divided into four quarters, alternating silver/white (argent) and green (vert).
Main charge (the creature):
- A hybrid beast statant affronté — A made-up creature standing on all fours (“statant”) and facing the viewer directly (“affronté”).
- Body and legs of a goose proper, spotted sable — The body/legs are those of a natural-colored goose, but with black spots.
- Head and elongated neck of a hound proper — The head and long neck are from a natural-colored hound/dog.
- Langued gules — Its tongue is red.
- Adorned with a Santa cap gules trimmed argent — It wears a red Santa hat with white trim.
In short:
A four-quartered white-and-green shield displaying a creature that has a goose’s spotted body and legs, the long-necked head of a dog, seen facing forward, with a red tongue and wearing a red-and-white Santa hat.
Or from a website that looks too old-fashioned to be AI:
Argent, (fr.): the tincture Silver. By those who emblazon according to the Planetary system it is represented by the Moon, just as the tincture of gold is represented by the Sun. Hence it is sometimes fancifully called Luna in the arms of princes, as also Pearl in those of peers. As silver soon becomes tarnished, it is generally represented in painting by white. In engraving it is known by the natural colour of the paper; and in tricking by the letter a. In the doubling of mantles it may be called white, because(as the old heralds say) it is not in that case to be taken for a metal, but the skin of a little beast called a Litvite. Sometimes, too, in old rolls of arms the term blanc is used.
Ooh, actually, Wikipedia has an article on this: Tincture (heraldry) - Wikipedia
Heraldric charges are designed to be descriptive enough to reproduce the image consistently, at least to a good-enough level. Any image that matches the description is considered an equally-valid image of the crest.
I don’t think that a trained human herald would ever use “a hound proper”, though, since hounds can come in so many different colors.
Sora has dropped from 30 free videos per day to 5.
Evidently they hit their adoption target. So now it’s time to drive away a bunch of your customers.
Sorry I didn’t read the whole thread, but I was thinking about this after seeing an ad that we frequently see recently on Amazon Prime TV. I forget what it is for - but it starts showing a woman stepping from an ocean onto a beach with incredible cliffs in the back ground. Then it changes to some people stepping out of a northern hot spring with snowy shores in the background. Then it changes to a blue green convertible driving down a winding road in some sunny landscape.
The first time I saw it, I commented to my wife on how beautiful and perfect the images were. Then, something made me think the first image was likely computer generated/enhanced. Then I thought the same of the next 2. And it sorta made me sad and disappointed.
Another ad features a man standing in a field of yellowish tea plants stretching into the background. I had the same reaction. FIrst, “How beautiful!” Then, “It is fake. A trick.”
Just a couple of TV ads that I don’t even pay enough attention to to say what they were for. But it just enhances my overall preference for the real over the computer generated/enhanced - whether images, film, writing, music…. I find myself increasingly appreciating what I perceive with my own senses - even tho it may be in my pedestrian suburban surrounding, over some image that I do not trust even reflects what occurred in nature somewhere else and someone else perceived with their senses.
I play in a 3-piece acoustic string band. Previously, I was somewhat critical of our efforts. When I listened to recordings, my ear would be drawn to whatever imperfections I perceived. But one benefit my perception of tech enhancements is that I feel better about our music. Whatever you may think about our music, good or bad, what you are hearing is our instruments and our voices. Generally un- or minimally amplified.
I often feel like a luddite. I do not wish to speak ill of all technology. That would be stupid. But I do not feel our existence is overall enhanced by the fact that it is increasingly difficult to discern between what is “real” and what is “artificial.”
When I saw this joke just as bare dialog over in the jokes thread, I thought, “That sounds like an xkcd comic” so I had to make it with ChatGPT:
The first try with just feeding it the dialog and telling it to make it an an xkcd style comic didn’t work out well and it spoiled things by showing the boiling water, so I got more specific on the second try for a perfect result.
Prompt
Create an xkcd-style comic strip starring black hat guy and cueball.
Panel 1:
Black hat guy: ‘The neighbor’s kids challenged us to a water fight.’
Cueball: ‘I’m in.’
Panel 2:
Black hat guy: ‘Just waiting for the water to boil.’
Cueball: ‘What?’
Do not show him boiling the water.
I think that having to spell it out that explicitly is proof that the AI does not, in fact, get it.
Apparently Sora dropped the 5 video limit almost as quick as they added it. I was there last night making Japanese gameshows and did more than a dozen of them without reaching a limit.
Bob Ross as a BL mangaka
Saw this image
Used “Stylish but illegal monkey found roaming Toronto Ikea” as a prompt in Sora and it got the outfit surprising accurate.
I noticed today Grok imagine video got a new update. It’s now doing unprompted zooms and cuts, and it takes emphatic prompting to get it to stop doing those. Before this, it was difficult to get it to do any cuts at all.
Interesting. I hadn’t noticed that before, but looking at recent Grok videos I am seeing heavy use of zoom, especially. Also, for the first time now I’m seeing a “rate limited” notice that I can’t make anything now between the peak hours of 8:30 AM and 1:00 PM (EST). I’ve hit the free limit before, but never with a specific notice like that. (And are those really those really the peak hours for Grok?)
I sprung for a month of supergrok so I am not seeing anything about rate limits. I see the reddit and discord for Grok are blowing up with complaints today; this new version is almost universally decried. Yeah, the characters are doing unwanted turns, movements, and staring right at the camera when before those issues were there but only occasionally. The zooms are now downright weird. Its prompt adherence has definitely gone downhill too.
Said another way, it knows better than you do what you want.
I’m now seeing that scene from many monster movies where the monster has been captured and restrained. But is now pulling at the cage or chains or whatever which are sloooowly stretching apart as everyone wonders whether they’ll hold. Hint: they won’t.
Which might itself be a fun prompt to feed into these systems.