I think VolvoHeritic is inventing the ekranoplan.
Realistic people are difficult. You can really tell the thing has been trained on Instagram, as it tends towards a certain generic facial type with a wide mouth and big cheekbones and a high forehead. Hardest part is getting it to put some texture in the skin and getting the model to keep her driving suit zipped. The Instagram influence also shows in the fact that it almost always draws NAs with the headlights up, and that's a relatively new thing.
Here's the best I've done. It helps that it's emulating old Kodachrome 64 film.
Movie ads for the movie 28 days later but it's about 28 days after Wisconsin runs out of cheese.
In reply to Keith Tanner :
Mind sharing the exact text you used? I find that the more words I use, the worse it gets. It doesn't help when I misspell them.
VolvoHeretic said:
In reply to Keith Tanner :
Mind sharing the exact text you used? I find that the more words I use, the worse it gets. It doesn't help when I misspell them.
I'm using an OpenAI subscription, but it runs Dall-e like Bing. The big difference is that I can iterate - if I want to change something, I can say something like "do it in X style" or "the car is parked on the beach". So it's hard to pull out the exact text. This should be pretty close, though:
A 1990 Miata with Martini Racing livery and a blonde female in a racing suit, showing excitement and slight nervousness, with both eyebrows lifted high as she looks at the camera, as if daring the viewer to race her. She has a mysterious smile. Her hair is curly and slightly messy. She is standing beside the car and leaning on the fender. Make it a photographic style, using a color palette characteristic of Kodachrome 64 film.
In reply to Keith Tanner :
I'd say you described and AI drew Farrah Fawcett.
The difference between Keith's AI program and the Bing AI program is like the difference between Married with Children and the Simpsons.
In reply to QuasiMofo (John Brown) :
Both are stupid but one's a cartoon?
QuasiMofo (John Brown) said:
The difference between Keith's AI program and the Bing AI program is like the difference between Married with Children and the Simpsons.
More like I've been able to twiddle the knobs more to learn how it works. I ran that prompt 5 times with 10 pictures as a result, I only showed you the best one :) And that was after I'd run it a bunch more times to get close.
Here's one that came out of Bing with that same prompt. Wide mouth, high forehead, a lot of color on the cheekbones and nose, strong jawline - same woman who shows up in a lot of mine. Still no wrinkles on the forehead. The overall image could have come out of either Bing or OpenAi, I get a lot that look like this.
This one from OpenAi was pretty good - slightly different prompt and without the Kodachrome. It's a well done illustration, but it was a one-off in the results. None of the others looked anything like this. Obviously, I'm using the same subject matter over and over to learn how it works. Either that or I don't have enough imagination to come up with sharks riding jetskis or whatever.
Aha, I just found out I can extract the actual constructed prompt used from OpenAi. Here's the "Farrah" pic from earlier.
A photographic style image emulating the color palette characteristic of Kodachrome 64 film of a 1990 Miata race car with Martini Racing livery, featuring blue, red, and light blue stripes on a white background. The car, true to its 1990 design, has distinct pop-up headlights and a classic body shape, and is parked on a race track. A blonde female with curly and slightly messy hair, wearing a professional racing suit, is standing beside the car and leaning on the front fender. She has both eyebrows lifted high as she looks at the camera with a mysterious smile. The rich, vibrant colors and fine grain texture of Kodachrome film enhance the scene's realism.
Drop that into Bing and you get...
Keith, I don't mean to steal your idea and I don't want no women messing up my car renderings, but I wanted to see what AI would do. It won't produce anything with a real name attached, no Farrah Fawcett or just Farrah after it new who I really meant.
Woman wearing a racing suit leaning on a road racing Ferrari 250 GTO with wide wheels and flares. Front aerodynamic splitter and rear downforce wing along with lower body skirting. Painted translucent candy apple red sprayed over brilliant white base coat.
No worries about "stealing" :) The whole blonde female beside a car thing is because my rally navigator wife is blonde so i try to see if I can steer the program closer to her. I wasn't actually aiming for Farrah, I'd just specified the curly blonde hair because I wanted to see if it would add to the realism. That just happened to be the one standout pic.
I think you can say you want someone who looks like a fictional character, such as "Jill Monroe as played by Farrah Fawcett in Charlie's Angels".
Specifying film or lenses has a fairly big effect on the result. So does specifying lighting. That's assuming you're after realism.
I see your female racing drivers have the same poor choices in footwear that mine usually do.
I tried your prompts and got basically the same thing in Bing Chat , and the same frizzy hair girl .
Did you extract the prompt from Bing or OpenAi ?
In reply to californiamilleghia :
I extracted it from OpenAi and ran it in Bing. My last post with pics were from Bing.
Most of my results look like that. The best one is an outlier and I think it was mostly due to the film treatment. Sometimes that gets ignored, sometimes not.
it does seem to ignore a lot of prompts , you get 4 images and they will all have one prompt or another ignored ,
Does OpenAi do the same ?
Yes. They're both running Dall-e. But it's quicker and easier to re-run the prompt in OpenAi. Plus you can give feedback with a thumbs up or thumbs down.
In reply to Keith Tanner :
I know nothing about lenses but newer AutoCAD knows what they are. Maybe I could get AI to quit cutting off the cars and fit the whole thing into the shot if I could give it a lens size and a focal length? I now just look at a view threw the camera in AutoCAD and move in and out, zoom in and out, pan left or right, move up or down, or look up or down to get the views I want.
Herbie the Love Bug race car lowered with wide wheels and fender flares racing modern race cars on rural California highways.
Of course, it forgot the other race cars or he's so far ahead that you can't see them.
My brother and I were picking on a friend about past vehicular exploits (semi legal).
Prompt: motor home racing at Isle of Man time trials
I mean if Hoopes had a beard, wore a MOFO shirt and had a cool FR-S...
My buddy's wife has characteristics that resemble a famous actress. I told Ai to add a scolding female in the style or likeness of her to his racing poster, it worked very well. You just have to find your famous doppelganger, a younger me could have used a young Toby Maquire. At least my mother thought so, I guess I trust her.
Or Pixlr has a face swap option.
The AI art seems to print pretty well once SaxyHero fixed it up a bit, here's the shirt she made for me.
I suck at picture taking too
Yeah the AI pics definitely translate well to t shirts