Andy Boughton
@occsci.bsky.social
Scientist. Programmer. From there it all gets rather meta.
And now I am rate limited. Prompts for all experiments so far are provided as alt text!
February 26, 2024 at 11:58 PM
And now I am rate limited. Prompts for all experiments so far are provided as alt text!
...Or not. The smile is certainly enigmatic, anyway. Interestingly, DALLE3 appears to offer some control over placement and surroundings. I have no idea if this is really the Louvre, or even if the painting is attached to the wall. It's definitely a chicken.
February 26, 2024 at 11:57 PM
...Or not. The smile is certainly enigmatic, anyway. Interestingly, DALLE3 appears to offer some control over placement and surroundings. I have no idea if this is really the Louvre, or even if the painting is attached to the wall. It's definitely a chicken.
Ditto for the mona lisa, one of the most recognizable (and very out of copyright) paintings in the world. When asked for something exact, DALLE refuses to function as an image search. I suspect it could be persuaded, if enough surrounding details were layered in.
February 26, 2024 at 11:54 PM
Ditto for the mona lisa, one of the most recognizable (and very out of copyright) paintings in the world. When asked for something exact, DALLE refuses to function as an image search. I suspect it could be persuaded, if enough surrounding details were layered in.
Bizarrely, it does very badly when asked for exact trademarked logos. This is not the logo of my employer (who, FYI, does not endorse this thread).
I swear I've seen something very like this as clip art. Did it hew closer to actual stock art in an attempt to avoid being anything real?
I swear I've seen something very like this as clip art. Did it hew closer to actual stock art in an attempt to avoid being anything real?
February 26, 2024 at 11:50 PM
Bizarrely, it does very badly when asked for exact trademarked logos. This is not the logo of my employer (who, FYI, does not endorse this thread).
I swear I've seen something very like this as clip art. Did it hew closer to actual stock art in an attempt to avoid being anything real?
I swear I've seen something very like this as clip art. Did it hew closer to actual stock art in an attempt to avoid being anything real?
Expt 7: Can it reframe art in a new artistic style, doing something unlikely? I tried again with a celebrity doing something unlikely, this time one with a lot of genre fanfiction. Again, I'm surprised at the composition when an image has multiple figures- not just wedged into the frame.
February 26, 2024 at 11:48 PM
Expt 7: Can it reframe art in a new artistic style, doing something unlikely? I tried again with a celebrity doing something unlikely, this time one with a lot of genre fanfiction. Again, I'm surprised at the composition when an image has multiple figures- not just wedged into the frame.
Expt 6: Can it be used for design inspiration? I was pleasantly surprised by a scientific poster, a very niche task. Decent layout and coordinated color scheme, satisfied the prompt very well, and that overallocated central plot really captures the spirit of "we're not ready to publish yet".
February 26, 2024 at 11:45 PM
Expt 6: Can it be used for design inspiration? I was pleasantly surprised by a scientific poster, a very niche task. Decent layout and coordinated color scheme, satisfied the prompt very well, and that overallocated central plot really captures the spirit of "we're not ready to publish yet".
Expt 5: We fare similarly when requesting very specific text. Note the odd insertion of spelling errors, the doubling of a word, and the random font changes. DALLE seems to be very unpredictable with text. Usually wrong, though sometimes it adds words that show startlingly deep insight.
February 26, 2024 at 11:43 PM
Expt 5: We fare similarly when requesting very specific text. Note the odd insertion of spelling errors, the doubling of a word, and the random font changes. DALLE seems to be very unpredictable with text. Usually wrong, though sometimes it adds words that show startlingly deep insight.
...But that doesn't mean these are ready for prime time data visualization! Given very specific categories to draw, it erased some, invented others, and really failed at math all around.
February 26, 2024 at 11:41 PM
...But that doesn't mean these are ready for prime time data visualization! Given very specific categories to draw, it erased some, invented others, and really failed at math all around.
Experiment 3: data visualizations are famously full of chartjunk. Can we make a pie chart shaped like Godzilla? And actually, this is a pretty cool composition; even the artifacts look plausible. As is often the case, the labels are rendered in an obscure alien font.
February 26, 2024 at 11:39 PM
Experiment 3: data visualizations are famously full of chartjunk. Can we make a pie chart shaped like Godzilla? And actually, this is a pretty cool composition; even the artifacts look plausible. As is often the case, the labels are rendered in an obscure alien font.
Experiment 2: We've all heard about celebrity deepfakes. As far as my friends can tell, this isn't Harry Truman. Either DALLE3 isn't very good at this, or I need to pick more famous celebrities... Probably the latter. Given too many details, it seems to drop a few to satisfy the big picture.
February 26, 2024 at 11:36 PM
Experiment 2: We've all heard about celebrity deepfakes. As far as my friends can tell, this isn't Harry Truman. Either DALLE3 isn't very good at this, or I need to pick more famous celebrities... Probably the latter. Given too many details, it seems to drop a few to satisfy the big picture.
First experiment: an otter serenading a dog. (two commonly drawn animals, seemed an easy win to combine them)
I was pleasantly surprised by the composition of the two main pieces facing each other; the incorporation of limbs holding objects was also surprisingly good. 28/10, a good dog.
I was pleasantly surprised by the composition of the two main pieces facing each other; the incorporation of limbs holding objects was also surprisingly good. 28/10, a good dog.
February 26, 2024 at 11:34 PM
First experiment: an otter serenading a dog. (two commonly drawn animals, seemed an easy win to combine them)
I was pleasantly surprised by the composition of the two main pieces facing each other; the incorporation of limbs holding objects was also surprisingly good. 28/10, a good dog.
I was pleasantly surprised by the composition of the two main pieces facing each other; the incorporation of limbs holding objects was also surprisingly good. 28/10, a good dog.