DID YOU FIND AN A.I. WITH A SPECIAL "DEFICIT"
I call it a misunderstanding mechanism。
It means:
You say one word, it's always understood another sentence。
MOST OF THE TIME, YOU CAN'T GET IT ON, NOT BECAUSE YOU'RE NOT GOOD ENOUGH, BUT BECAUSE AI'S SMART。
For example, when you don't write anything, it thinks you want to "default the red face of the net," and you write "fantasy," and it starts to smoke。
To make it real, we can't follow it. We have to do the opposite。
Today's lesson, I'm taking you to the secret of “reverse control” by defusing these three misunderstandings。
First big misunderstanding: Misread the word "default" to "default effect"
Principle Analysis: What is the default word
THE ONES YOU DIDN'T WRITE, BUT AI FELT "MUST HAVE." AI'S TRAINING DATA ARE FULL OF LATE-PROCESSED IMAGES。
So:
WHEN YOU WANT A CLEAN PICTURE, AI THINKS YOU WANT A "RICH DETAIL TEXTURE" AND THEN YOU GET A LITTLE NOISE
WHEN YOU WANT THE CHARACTER TO BE NATURAL, AI WILL THINK YOU WANT A HIGH-SOFT SUGAR TABLET, AND YOU'LL GET A SHARP FILTER。
And the result is, the more you think about nature, the more fake it is。
Operational solutions: a clear “prohibition order”
THE SOLUTION IS VERY SIMPLE AND ROUGH: JUST TELL AI WE DON'T WANT ANYTHING。
The more clearly you forbid it, the more afraid it will be。
In MidJourney, we can use the --no argument (negative hint):
Scene One: Rejecting “plastic sharpening”
We don't write Ultra realistic, because it would drive AI crazy。
Photo of Pepe the Frog holding black standing in each, holding a red rose, against a sky blue base, with bright pastel colors, in the stage of Martin Parr
"Soft natural skin texture (soft and natural skin texture) with negative: –no over-sharping, plain skin, heavy contrast"
With the stone of Martin Parr
Scene 2: Rejected "dirty texture"
Clement: Clean minimural background
–no negative, noise, chaotic details
the sky is a clear blue, and there are mountains in the distancing
THINKING: WHEN YOU SAY "DON'T SHARPEN," AI'S ONLY GOING TO LOOK FOR RAW, UNMODIFIED DATA, SO YOU GET A "REAL PICTURE" OF THE FILM。
Second big misunderstanding: Misreading the word “abstract” as “exaggerated special effects”
IT'S THE EASIEST PIT FOR A ROOKIE. AI UNDERSTANDS ABSTRACT ADJECTIVES VERY SUPERFICIAL AND RIGIDLY。
You write Dreamy -- AI starts painting smoke there。
You write "high-end" -- AI starts to give you plaster and gold powder。
Operational solutions:
Create Visual Dismantling Formula
WE'RE GOING TO BREAK ABSTRACT WORDS INTO QUANTIFIABLE VISUAL ACTION. IN OTHER WORDS, AI REALLY UNDERSTANDS WHAT YOU MEAN。
Here are three sets of the master formulas that I have summed up:
Formula 1: Dreamy, don't write Dreamy, write:
Low saturation (low saturation) + Soft diffused light (soft and flashing) + Telephoto lets (long focus)
she turned back to us as a low-faces away from the Camera and held up with light flying through the leaves to her skin
Formula 2: High-end Do not write High-end, write:
Strong light contrast (light contrast) + Negative space (white) + Clean command
three lotus leaves, long blues back in the ground, clear conditions, high-level relationship, multi-widge-angle lens, soft lighting, statistical relations of lous leaf settlements
Formula 3: Texture. Do not write Texture, write:
Subtle sidelighting
called in light, calling your name, nostalgic, non-smally, analog turns, phone call, non-physical atmosis, soft pinkglow, melancholy, retro sunset, black lightlights, cabina, soft funcus, pastel color, left secretes, flying screech, one-sided screechette, one-sided atmosher, one-sided, one-by-carl lorcher, one-by-Alex Grey, one-by-Greey, one-by-man-man, one-by-man-man, one-man-man, one-man, one-in-man, one-in-man, one-in-man, one-in-man, one-in-man, one-in-man, one-in-man-in-man, one-in-man-in-man, one-in-man-in-man, one-in-person, one-in-one-in-man, one-in-in-one-in-one-one, one-one-
Third misunderstanding: Misunderstanding the “uncertainty description” as “automatic completion”
AI HAS A “COMPULSORY DISEASE” THAT CAN'T BE SEEN IN THE PICTURE。
IF YOU HAVE AN INACCURACIES, AN "UNCERTAINTY" PLACE, AI ACTIVATES THE "AUTOMATED COMPLETION MECHANISM."。
It's a default to help you with the background, the environment, the road people, and it's always going to make your whole picture taste different。
Practical solutions: precision limits, a prison
THE MORE RESTRICTIVE YOU ARE, THE MORE I'M AFRAID TO PAINT. YOU'RE GOING TO BE THE CIRCLE。
A car in street
AI HEART ACTION:
Where is the street? Is there anyone there? The town? The village。
Correct (lock border):
a clinical vertical woman in the early 1980s standing behind a classy young Jacob E-Type Parked in front of a brownstone building in Manhattan, she wears sunglases and a winding pastel drink, the car gleams, replays on its plumes, soft shadows, subtle Wessons most highly, symmery in colour palette,
A graphic image of a young fashion woman in the early 1980s standing next to a classic yellow panther E-Type parked in front of Manhattan's Palm Building. She wears sunglasses, she wears a light dress, she shines in the light of golden times, the car flashes, chromium plating parts reflect light, soft shadows, a fine Wes Anderson style, symmetrical layout, retrospect
When you lock the background:
standing behind a customary Jaguar E-Type Parked in front of a brownstone building in Manhattan
AI CAN'T "AUTOMATE" THE TREES AND ROAD PEOPLE THERE。
End of story: summary of heart
WE'RE ACTUALLY LEARNING THE ART OF COMMUNICATION. AI IS LIKE A SKILLED PAINTER WHO DOESN'T KNOW HOW TO LIVE。
It's self-advocacy, so we have to know to reject。
It has limited capacity (in abstract words), so we have to dismantle the command (visual formula)。
It's idle (auto-fill), so we have to set up borders (precision limits)。
In conclusion, I would like to say that the good words are not to be completed at once, but need to be adjusted at once。
YOU CAN READ YOUR MEANING BY OBSERVING THE MISPERCEPTIONS OF AI AND ADJUSTING YOUR LANGUAGE。
PATIENCE, ACTUALLY, IS AN IMPORTANT PART OF AI'S CREATION。
Attention to the subject:
Software tool: MidJourney https://www.midjourney.com