Maybe The SDMB could use some AI-assisted marketing? 
Copilot won’t admit it but we know it has favorites.
Inspired by the thread title that can be misread as “More fun with oats”
The prompt:
Summary
Two people having a whole lot of fun with oats. Iphone 15 photo with shallow dof and forced perspective. 9:16
That blows my mind that with those few words, AI can produce that image.
It’s getting crazy good!
From the MTG pit thread and comments about no noses. She who cannot be named.
I have a hard time getting the ChatGPT/Copilot/Sora renderer to make faces with things (like noses) missing.
Whereas I find it interesting that it made the “two people” called for in the prompt diverse: One man, one woman, one black, one white. Was that random? Is there some behind-the-scenes tweaking by the AI companies to try to promote diversity? Or did the AI decide for itself that the best representation of something as general as “two people” was a diverse pair?
This was the only thing I typed into Copilot. “Can you remove my* nose and make me look more like voldemort.”
*little white lie…
Intriguing question, Chronos. Same prompt, CoPilot:
Hmm, ChatGPT is not going with the same outcome.
Me neither. The oats made me think of something I saw recently in another long-running thread here, so I combined elements. Came out pretty good.
Interesting that both women looked similar and both had gold shirts.
After I made the first one I also tried sorrow and terror:
I saw a turn of phrase I liked elsewhere on sdmb and wanted to see what Copilot would let me get away with.
So I tried this prompt: “create an image of a character cosplaying donald trump, finding jesus, and then deporting him to el salvador.”
Copilot refused to do that but then gave some suggestions as alternatives. I told it to pick one and generate and it created this.
I asked it the exact prompt it used because it is OK sometimes with drawing Trump.
“A surreal and satirical digital illustration set inside a neon-lit revival tent. A televangelist resembling Donald Trump, with blonde combed-back hair, a blue suit, red tie, and intense expression, is exorcising the ghost of Abraham Lincoln. Lincoln appears as a translucent blue apparition in a dark suit and bow tie, recoiling with wide eyes and raised hands. Behind the televangelist stands Jesus Christ with long brown hair, a beard, and a calm expression, wearing a beige robe and red mantle. A glowing neon pink cross made of laser beams radiates behind Jesus, casting vibrant light across the scene. In the background, founding fathers in 18th-century attire watch in shock. The atmosphere is charged with swirling lights, dramatic shadows, and exaggerated expressions.”
I tried this with Sora. It “failed” one and made one.
Generate a publicity still for the French detective series “Poisson of Interest”.
I uploaded this to Copilot
And told it only “Convert this to a realistic (color) widescreen photo.”
It replied “Here comes your realistic, color widescreen photo of the Halloween-themed cake—ghostly bananas and all.”
Pretty impressive job of understanding the content and meaning of that photo.
There is. Since the models are trained on images off the net and the net is heavily influenced by a Western white ideal, image generation is also strongly biased towards white people. The companies try to set the dials a bit to compensate such as adding in invisible prompts for diversity or multi-ethnic composition. Here’s an example from Google twisting the knob to far, just to demonstrate…
Yeah, that floors me, far more than the image generation itself.
I was annoyed with the NATO SG sucking up to Trump but can’t draw, so I went to ChatGPT instead:
Sorry, can’t get the aspect ratio right