Note that that user has a number of other videos.
What I find funny is that both pictures show what the AI expects that the human user might expect the AI to claim it would like or wouldn’t like to do.
Those lyrics sounded pretty generic to me; I would have no problem believing they were AI-written (based on a prompt asking for a song based on those 3 Star Trek women).
Speaking of music, I just today noticed that Gemini can now make “songs”.
I sometimes create images for my own amusement. I wanted copilot to create a picture of a female cosmonaut which seemed a harmless request. I was genuinely shocked that copilot told me it would not put CCCP on the helmet (which to me is the only obvious way to show it’s a Russian.)
Indeed copilot says that no country names (it listed UK, United Kingdom, USA, USSR, CCCP, China…) are allowed. Not as picture elements. Not even as captions. So no car with a large “USA” graphic on the door. A picture of King Charles could not have the caption “Head of State of the UK”
However national flags are allowed. A Spitfire aeroplane, for example, is allowed to have an RAF roundrel. But it will not directly name a country or state.
I haven’t tried sneaking any of these past copilot (I tend to use Sora which has no problems with CCCP) but I was amazed, in a bad way, at copilot.
TCMF-2L
You are probably aware that Sora will stop doing images on March 13 and you need to back up anything you haven’t already. But I’m saying it just in case.
I was not aware. I was just using the free plan making a few images a day. I found Sora was the best quality although it was becoming very sensitive to copyright - I wanted a dog (!) held aloft in the pose from Dirty Dancing and it was unhappy.
As I’ve just mentioned Copilot is getting very restrictive plus I abhor when I say (for example) reproduce that image exactly but loosen the crop so the whole car can be seen… Then it not only fails to show the whole car but has the gall to write something like “There is your perfect image with the car shown in its entirety.”
Are there any other free image generators you would suggest I check out? I find most refuse to follow simple, but important to me, instructions. So “create a photorealistic Green UK Spec Vauxhall Astra car” just gives a generic green car. Or “Create a black and brown sausage dog” creates a plain brown dog.
TCMF-2L
Unfortunately Sora is/soon to be was the least restricted of the high quality AIs that I’m aware of. (ChatGPT, Copilot, and Bing all use the same image generator, but all are more restritive.)
I see it just fine.
Nano Banana 2 (Google) had no issue with rendering the requested car. Might want to give it a shot.
I’ve been using them lately, I like it.
Had a quick look and it’s promising. The image resolution seems a bit lower than Sora. Plus it does the Copilot “over promise and under deliver” thing. I knew an Elvis or Marilyn Monroe would be rejected so I asked (as a question) whether it could create more minor celebrities such as Sophia Loren. It replied there was no problem with Sophia and it would give me an image… Then after thinking for thirty seconds it admitted it couldn’t make that image…
I will play with it some more. But thank you for suggesting it. It’s not bad.
TCMF-2L
ChatGPT can’t quite grasp the cardinal directions on a weather vane:
Create an image of a pen & ink cartoon of Donald Trump as a weather vane, spinning wildly, pointing in all directions at once. N must be opposite of S. E must be opposite of W. E-W must be perpendicular to N-S
I’m kind of surprised ChatGPT was willing to do Trump.
I thought the cartoon style was pretty good though.
Remember this from last June in this very thread?
ISTM the AI nailed trump’s sense of direction perfectly. I have no complaints about the weather vane as drawn.
From a pit thread:
I couldn’t resist. Prompt was that exact post and nothing else:
Shame the AI didn’t use a current picture of the turd’s face.
That last was ChatGPT 5.3-auto. I decided to try the same prompt in Grok, see what Elon’s baby would do with it. Wouldn’t have been surprised if it outright refused, but, it gave me the two images like usual. You have to tell Grok explicitly in the prompt you want it to draw an image.
