When AI Goes Wrong

[john]

Hi

 

Inspired by a topic by @GregB if you’ve been using AI and it has gone spectacularly wrong, for creating images, video, music, lyrics etc…. Post up your bad results!

 

 

cheers

 

John

[GregB]

TWO recent examples ... 

 

a) medieval longbow archer, photo, realistic, --ar 16:9

THese are just 4 out of probably 36 ... I've been unable to get ANY completely logical archery images.

 

b) using ‘Vary Region’ to replace perfectly presented  medieval knights on horseback with ‘longbowmen archers and men-at-arms’ ...

... WTF?!  I was literally stunned and slack-jawed for several seconds when this appeared.

[GregB]

What I had was not bad ...

column of 15th century knights on horseback on open country dirt road viewed from the rear, photo realistic, --no crosses spears --ar 16:9 --style raw

 

A small tweak was intended using this VARIATION  ...

close up of long loose column of 15th century medieval knights on horseback galloping away on country dirt road in open countryside, flowing horses tails, viewed from the rear, photo, realistic, --no spears --style raw --ar 16:9

Great for Bollywood, but nowhere near what I wanted ... or expected!!

[john]

51 minutes ago, GregB said:

What I had was not bad ...

column of 15th century knights on horseback on open country dirt road viewed from the rear, photo realistic, --no crosses spears --ar 16:9 --style raw

 

A small tweak was intended using this VARIATION  ...

close up of long loose column of 15th century medieval knights on horseback galloping away on country dirt road in open countryside, flowing horses tails, viewed from the rear, photo, realistic, --no spears --style raw --ar 16:9

Great for Bollywood, but nowhere near what I wanted ... or expected!!

 

"...a cheer went up as a chapter of Hell's Knights hove into view..." lol

[GregB]

Wowsers! 

Seemed pretty simple ... close-up of a kind smiling face , realistic --style raw --ar 16:9 

BUT ... (spot the mistake!)

 

[john]

Big hairy lips?

 

oh, you mean the eye staring out of a shimmering planet in the middle of the upper teeth huh? Sheesh, some people are picky! Lol

[GregB]

The very new and free Dall-E 3 has been highly praised for it's ability to create words within images.  Prior to this, Dall-E - and also others like MidJourney - could only display gooble-de-gook scribbles.
 

My prompt was:

photo of a pretty blond woman in her twenties wearing a white t-shirt which says "The Flat White Album" written in large black capital letters.

 

The resulting lettering IS remarkably good and it follows contours of both the body AND the cloth folds, but not one of the three logos presented was spelled correctly!!

No doubt this will be fixed within weeks ... and then MidJourney will likely frog leap Dall-E with many new features.

 

Progress in AI image generation is absolutely astounding.