A bit of fun with AI
- Thread starter FretfulMass
- Start date
North Hampshire
Try Microsoft Copilot, and just enter a statement like, "Create an image of a BMW Z4 relating to drag racing". For images you have to have an account, but that is easy to setup.
Attleborough, Norfolk
Ohio
With some training, you can have some fun with it....
It does appear a bit challenging for the AI to put a driver actually in the car properly, i tried a few times with no good results as yet
Scotland 🏴
south yorkshire
Much better, this one was done using ChatGPT Sora
@paddy wright If only such leggo existed, nice touch with the number plate 
AI can be fun at times. If anyone wants to play, I like using gemini by google. (gemini.google.com)
Just go here and tell it what you want in the image, and keep making adjustments until it's what you want.
I asked for a Z4 being driven by a blind man and got this:
Then I decided that I liked the idea of a dog driving and asked to remove the human and got this:
Then I thought the dog looked too big for the car, I asked for a smaller dog and got a smaller breed
:
Then I asked it to fit the dog inside the car and got this:
Just go here and tell it what you want in the image, and keep making adjustments until it's what you want.
I asked for a Z4 being driven by a blind man and got this:
Then I decided that I liked the idea of a dog driving and asked to remove the human and got this:
Then I thought the dog looked too big for the car, I asked for a smaller dog and got a smaller breed
:Then I asked it to fit the dog inside the car and got this:
Funny how quickly these tools turn into a kind of iterative creative loop.AI can be fun at times. If anyone wants to play, I like using gemini by google. (gemini.google.com)
Just go here and tell it what you want in the image, and keep making adjustments until it's what you want.
I asked for a Z4 being driven by a blind man and got this:
View attachment 282758
Then I decided that I liked the idea of a dog driving and asked to remove the human and got this:
View attachment 282759
Then I thought the dog looked too big for the car, I asked for a smaller dog and got a smaller breed
View attachment 282760
Then I asked it to fit the dog inside the car and got this:
View attachment 282761
I’ve seen a similar workflow people use with Atomic Chat as well, not for images specifically but for running different AI models locally and iterating on prompts without relying on cloud tools. It’s a different angle, but the same idea: fast feedback loops and incremental refinement until you land on something useful.
What you’re basically doing there is closer to “directing” than generating — each prompt becomes a small edit instruction, and the model acts like a fast visualization assistant. The interesting part is how ideas evolve on the fly: blind driver → absurd concept → refinement → correction of scale and physics → final composition that feels “right enough”.
This kind of workflow is actually where image models shine the most — not in the first output, but in that back-and-forth shaping process. It’s almost like sketching with a very patient (and slightly literal) illustrator.
One thing I’ve noticed though: the more specific you get about spatial relationships (“dog must fit properly in seat”, “scale relative to steering wheel”), the more stable the results become. Otherwise the model tends to optimize for concept over realism.
Curious what you’d get if you pushed it further — like changing perspective (outside shot vs interior camera), or switching from realism to stylized render.
Chester
I would buy one of these!