Whooo’s a Lawful Good doggy? You are! Yes, you are!
Here’s your lawful-good Doggie:
It’s interesting scrolling through this thread. The demarcation from being impressed with anyone getting a symmetrical face to it being no big deal is quite sharp.
Early in the thread I commented about how exciting it is to experience the beginning of a new, disruptive technological advance. Every time I start to get burned out, some new advance is made that catches my attention all over again.
Speaking of attention, the word is getting around about the Bing upgrade. Not only are the GPUs being overwhelmed right now, the web interface itself is barely functioning.
Yeah, I thought I’d give it a go for the first time. It said it would take 15 minutes to create, but after letting it run for an hour and a half with nothing produced, I gave up.
Under normal conditions it takes well under a minute. I generated around 70 sets of images today “normally” before the traffic started to hit.
I’ve had it running for an hour now with no result. No big deal since it’s just a browser tab but I’m not optimistic.
Microsoft is bringing “thousands” more GPUs into the mix.
https://www.reddit.com/r/dalle2/comments/16x9fiu/tweet_from_microsofts_mikhail_parakhin_folks_we/
I just had to have Bing create a feline version:
I also liked the Action Figure Zappa prompt by @Walken_After_Midnight . I had Bing create action figures of the band My Chemical Romance, Photoshopped the bad text, and sent them to my daughter, who is a fan of theirs. Now I feel bad…she wanted to buy them.
All in all, I’m pretty impressed with Bing’s AI. Its graphics are pretty close to what you prompt (even the text, for the most part), and the images are rendered almost as highly detailed as Midjourney’s.
And, according to at least one tutorial I watched, Bing’s text/language AI on Creative Mode uses version 4 of ChatGPT. Not sure if that’s true, but the output is spot on. I’m a fan.
SDXL can do action figures, too. I did these at Playground a week or two ago (prompt: super cat action figure in box).
https://i.postimg.cc/JnZqgsqP/super-cat-action-figure-in-box-grid.jpg
Probably not long from now, the AI will be able to make a 3D version of the request… which you’ll be able to send to a 3D printing company, and get back a physical copy in a day or two.
The new Bing creator is great. “a black cat curling up on a white chair, cozy room, sunbeam shining in, softly lit, pastel, hand drawn, illustrated”
“Cecil Adams from the Straight Dope website”
“Action figure of Cecil Adams from the Straight Dope website”
“Action figure in a box of Cecil Adams from the Straight Dope website”
It wouldn’t allow “Photo of Cecil Adams from the Straight Dope website” as it said it violated terms.
“All the posters from the Straight Dope message board in one picture”
I’ve been trying complicated prompts to “stress test” DE3 and one prompt was “A photo of a chimpanzee dressed like Vincent van Gogh holding a painting of a giraffe by Monet” and was very surprised when a very close reproduction of Starry Night showed up in one of them.
Here are the other three from the set. All four managed a photographic image of a chimp holding a painting of something, but all paintings are in the style of van Gogh, not Monet. As for clothing, two got generic “artist berets” but the Starry Night one got the yellow straw hat I was hoping for. None of the images are completely perfect, but all are impressive.
I find a lot of what I want to create gets disallowed, such as:
“Elon Musk fighting an ostrich”
“Elon Musk fighting an ostrich and getting disembowelled”
“a man that looks like Elon Musk fighting an ostrich”
“Putin villain action figure in box”
Ed’s a bit less illusive than Cecil. I tried Ed with Bing, and got closer, but still had to use Inswapper in Midjourney (and minor Photoshopping) for a more faithful rendition:
Another “Cecil Adams from the Straight Dope website” appeared in my history. Same prompt as previously, but different pictures produced. Looks more like the world’s smartest human.
Bing restricts you from using names of prominent living celebrities. It doesn’t seem to care about most dead ones. For instance, on the new system I’ve successfully created images involving Elvis and Marilyn Monroe. I’ve made a number of Bob Ross painting something. But I couldn’t make Bob Ross painting Taylor Swift riding a bicycle.
(But you can get around blocks sometimes. Want an actor? Use the name of a character they play. Want Musk? Maybe asking for “Tesla CEO” will work.)
Yes, the limitations and restrictions are frustrating. One of my clients sells hand and foot products. Firefly seems to believe bare feet are pornographic and I get scolded when I try to render product shots that include a foot. With other AI programs that don’t restrict as much, it usually takes many variations until I get something that doesn’t look like the hands and feet of a mutant monster—not a good look for trying to sell foot creams!