In contactAI VideoHave you ever experienced such a situation during your generation:
IT'S A HUNDREDS OF WORDS-OF-WORD TIP, AND IT'S ALL ABOUT WHAT YOU CAN THINK OF, BUT THE IMAGE THAT AI GENERATED IS COMPLETELY OFF YOUR EXPECTATIONS
ACTUALLY, IT'S NOT THAT AI'S NOT SMART, IT'S BECAUSE MOST OF OUR HINTS ARE NOT VALID. IT SEEMS TO PROVIDE A GOOD AMOUNT OF INFORMATION, BUT THE VISUALIZATION LOGIC AT THE BOTTOM OF AI SIMPLY DOES NOT UNDERSTAND THE DESCRIPTION OF “HUMAN PERSPECTIVE”。
TODAY, BY DECONSTRUCTING THE THREE CORE ERRORS THAT YOU MOST FREQUENTLY MAKE IN YOUR DAY-TO-DAY OPERATIONS, WE WILL TAKE YOU BACK TO THE BOTTOM LINE OF COMMUNICATION WITH AI. WITH THESE THREE METHODS IN MIND, YOUR HINT WILL BE UPGRADED FROM “BLIND LUCK” TO “PRECISION CONTROL”。
Error one: A large pile of useless hints that ignore the initial frame weights and visual references
MANY PEOPLE ROUTINELY MOVE THE LOGIC OF TEXT-GENERATED IMAGES TO DESCRIBE THE IMAGES IN A LOT OF STYLE WORDS WHEN THEY DO THE “TURSON VIDEO”. FOR EXAMPLE, THE PHRASE "OVERWRITTEN, FILM-CLASS PHOTOCOPY, 8K RESOLUTION, HIGH-QUALITY" IS WRITTEN IN THE HINTS。
BUT IN FACT, IN THE CURRENT BOTTOM LOGIC OF THE TOP-OF-THE-ART VIDEO, THE FIRST FRAME IS “ABSOLUTELY HEGEMONIC”。
This frame has completely determined the visual style of your video, the optic structure and the underlying details. Blindly placing style words in video alerts is not only useless, but also severely dilutes your “action weight”。
Redundancy of style words and “calculation waste”
In the bottom logic of the Tusheng video, the highest weight is always your first frame (the original image you uploaded). This frame has completely determined the visual style of your video, the optic structure and the underlying details。
In order to give you an intuitive understanding, we're looking at a typical original test:
The image is of a girl in a dark coat, standing in a cold, cold, ice-free land, with a very pure solar-tide blue and blue。
midjourney V8.1 Generated, photoprompt:
A Chinese girl in a dark winter school stands on a dangerous scale.
WHEN WE FEED THIS TO AI'S BIG VIDEO MODEL AND FORCE INTO CONFLICTING STYLE WORDS, LOOK AT THE BOTTOM OF THE LOGIC GAME。
"Treatful Case"
THE WRONG HINT: "THE SABERPUNK LIGHT, UE5 TOP LEVEL 3D RENDERING, EXTREME CLARITY, SUPER HIGH QUALITY, GIRLS SWING AROUND SLOWLY, AND FINALLY LOOKING AT THE CAMERA."
A LOT OF PEOPLE THINK THAT THE ADDITION OF CYBERPUNK, UE5 WOULD MAKE THE PICTURE EVEN COOLER. BUT WHEN IT'S ACTUALLY PRODUCED, YOU'LL FIND THAT IT'S NOT EVEN CYBERPUNK
WHY? BECAUSE NOW AI IS VERY SMART, IT FINDS A SERIOUS CONFLICT BETWEEN YOUR WORDS AND THE ORIGINAL IMAGES UPLOADED. IN ORDER TO PRESERVE THE LEGITIMACY OF THE ORIGINAL MAP, AI WILL DIRECTLY QUALIFY THE TERMS "SABUBUNK, UE5, SUPER-HIGH" AS "INNOCENT NOISE" AND BLOCK IT
(However, this type of picture is not intended to include the phrase “Sebbunk”, mainly for case purposes。
THIS MEANS THAT THE CALCULUS WEIGHT OF 801 TP3T HAS BEEN WASTED。
Positive cases
Correct hint:
The girl turned in the snow and looked naturally to the camera. In the moment of turning, the cold wind lightened her hair and dark cloaks。
Parsing:
DELETE COMPLETELY ALL RENDERING, PAINTING AND STYLE WORDS. LET AI UNDERSTAND THAT ITS TASK IS SIMPLY TO DRIVE PHYSICAL MOTION, NOT TO RESET VISION. WE'RE GOING TO SAVE ALL THE LEAD WORD WEIGHTS, ALL OF WHICH ARE USED TO DESCRIBE PHYSICAL DETAILS SUCH AS “OBSERVATION” AND “BLOWING UP”. NOT ONLY DOES THE RESULTING VIDEO KEEP THE ORIGINAL IMAGE PERFECTLY COOL, BUT THE PERSON'S DYNAMICS BECOME EXTREMELY SILKY AND DELICATE。
2. Validation of numerical category directives
IN ADDITION TO STYLE WORDS, ANOTHER COMMON INVALID PILE IS " ABSOLUTE VALUES " . AI DOES NOT HAVE THE CONCEPT OF A REAL PHYSICAL WORLD MEASURE, AND IT DOES NOT UNDERSTAND EXACTLY HOW MUCH PIXELS ARE MOVED IN THE CURRENT IMAGE SCALE BY “SNOW 50 MM” OR “MOBILE 5 M”。
[Practical Case Presentation]
Initial state: still using this picture of the girl in the ice desert。
[Reverse case]
The wrong hint: "The wind reaches level six, it's very large. The girl moves forward at 1 metre per second, walking about 5 metres away and looking sad.”
THE PHYSICAL VALUES “WIND POWER LEVEL 6”, “1 M PER SECOND” AND “5 M RANGE” CANNOT BE ACCURATELY CONVERTED。
[Face Case]
Correct hint:
“The cold wind of the twilight blows, and the hair and tie of the girl is blown up in the back. Large snowflakes continued to float on her dark coat shoulder, and she slightly bowed her head and breathed a clear white fog.”
Parsing:
ABANDONING THE SPECIFIC WIND LEVEL TO REFLECT THE WIND BY “THE HORNS ARE BLOWN UP” AND ABANDONING THE SPECIFIC DISTANCE OF MOVEMENT BY “A STEP FORWARD” AND “A CLEAR FOOTPRINT” TO PROVIDE A PERFECT PHYSICAL SPACE REFERENCE FOR AI. THAT'S HOW THE PERSON'S MOVE IS SOLID AND HAS A REAL SENSE OF WEIGHT。
FOCUS ON THE INTERACTION OF THE DETAILS OF THE UPPER HALF. THE FINE PHYSICAL CHANGES OF “MICRO-LOW HEAD” AND “OUT OF WHITE FOG” PROVIDE A PERFECT VISUAL REFERENCE FOR AI. NOT ONLY DOES IT FIT PERFECTLY WITH THE ORIGINAL PICTURE, BUT IT ALSO FILLS THE PICTURE WITH A REAL SENSE OF BREATHING AND FILM。
ERROR TWO: THE HINTS CONFLICT WITH CONTENT, LEADING TO A LOGICAL SWING IN AI
Many of them appear to be very professional, full of high-level instructions, but in fact they are full of conflicting conflicts. This often happens when you are used to using long text templates。
The most typical example is the fact that many people have written “hand-held lenses” and “extremely smooth film-grade slides”。
20 Mirror Tips
WE HAVE TO UNDERSTAND THAT THE AVM IS VERY SMART. WHEN IT CONFRONTS THIS COMPLETELY CONFRONTATIONAL ORDER, IT DOES NOT “MEMMELT” OR “SHAPE” AS IT DID IN THE EARLY AI, AND ITS BOTTOM-UP TREATMENT MECHANISM BECOMES: FORCED PEACE OR OUTRIGHT NEGLECT。
SINCE YOU WANT TO “SWAY” AND “SOFT”, AI SIMPLY OFFSETS THE TWO INSTRUCTIONS AND THEN CREATES YOU A VERY MEDIOCRE, STUPENDOUS, SLOW SHOT LIKE PPT; OR WHEN IT DISCOVERS A SERIOUS CONFLICT BETWEEN THE “HIGH CONTRAST STUDIO” YOU WROTE AND THE “SOFT AND NATURAL LIGHT” ON THE ORIGINAL MAP, IN ORDER TO ENSURE THE LEGITIMACY OF THE PICTURE, IT WILL REDUCE YOUR LIGHT VOCABULARY DIRECTLY TO ZERO。
And the end result is, you wrote hundreds of words of advanced vocabulary, but you completely lost control of the picture. So let's figure out who's in charge of this shot, emotional or advertising before we write a hint
We'll take this indoor book as a case in point。
midjourney v. 8.1 generation, prompt:
Onematic shot, Shunji Iwai's “Love Letter” moved above histic,
Thumbnail video control (loss of control as a result of the presentation of the "Professional Glossary " )
As we enter the video generation phase, we look at the real conflict instructions that many people write in pursuit of "film."。
[Reverse case]
The wrong hint: "The film-grade mirror, with a strong shoulder-carrying lens, displays emotional tension, while the lens moves forward slowly and very smoothly. The very dramatic, high-comparison studio was empty, and the girls looked at it in silence.”
THIS HINT BRINGS TOGETHER ONE OF THE HOTTEST WORDS ON THE INTERNET, BUT IN FRONT OF AI NOW IT IS AN "INVALID COMMAND":
1. The mirror offsets each other:
THE TRAJECTORIES OF “HAND-HELD SHAKES” AND “SOFT THRUSTS” ARE THE OPPOSITE. POWERFUL AI WILL NOT LET THE PICTURE COLLAPSE, BUT IN ORDER TO BALANCE THE TWO WORDS, IT WILL ULTIMATELY GIVE ONLY AN INERT, SLOW AND RIGID PACING. THE EMOTIONAL TENSION YOU WANT IS TOTALLY IMPOSSIBLE。
2. The Shadow Command is ignored:
WHILE THE ORIGINAL MAP IS “SWEET, NATURAL REFLECTIONS OF THE DARK DAY”, YOU ARE FORCED TO ADD “SHOW OUT THE STUDIO”. NOW AI, IN ORDER TO MAINTAIN THE PHYSICAL LEGITIMACY OF THE INITIAL FRAME, WILL DIRECTLY BLOCK YOUR LIGHT COMMAND. YOU WASTED YOUR MATH AND CHARACTER, BUT YOU DIDN'T DO ANYTHING。
[Face cases] (right optimization: use of subtraction to establish a single dominant logic of silence)
Correct hint:
"Standing position. To maintain the light as soft and natural. The girl's finger turned softly on the next page of the book and her eyes remained focused on the book.”
Parsing:
1. Unified visual master logic:
SINCE THE TONE OF THE ORIGINAL FIGURE IS “QUIET, INNER CONDENSED”, WE HAVE ESTABLISHED “QUIET” AS THE DOMINANT LOGIC. DECISIVELY DELETE ALL VERBS THAT CAN CAUSE INTERFERENCE, SUCH AS SHAKING, DE-ORBITING, AND USE “FIXED POSITION” DIRECTLY AND EXPLICITLY. ONCE AI HAS RECEIVED A CLEAR AND SOLE DIRECTIVE, IMPLEMENTATION WILL BE EXTREMELY PRECISE。
2. Do not interfere with the light, focus on action:
TELL AI EXPRESSLY TO “MAINTAIN THE LIGHT” AND LEAVE ALL THE HINT WEIGHTS TO THE ACTION OF “TURNING THE NEXT PAGE OF THE BOOK”. INSTEAD OF MAKING UP CRAP, AI CAN MAKE THE DETAILS OF THE ACTION EXTREMELY REAL。
Error three: Too many commands are given at the same time, leading to an unlimited dilution of the weight of sports
Many creators are used to putting all action instructions in one sentence when they write a hint: things that want people to run, to turn the camera, and to keep the background flying。
NOW, THE TOP-OF-THE-ART AI MODEL, THOUGH POWERFUL, DOES NOT EASILY BREAK THE PICTURE, BUT WHEN YOU OVERLOAD, AI BECOMES VERY "FALSE." IT WOULD, IN ORDER TO BALANCE ALL THE DEMANDS, IMPOSE A LEVELLING OF THE WEIGHT OF THE EXERCISE, WHICH WOULD HAVE RESULTED IN A VERY MEDIOCRE RESULT: THE CAMERA WOULD HAVE JUST SWAYED, THE PERSON WOULD HAVE DONE ONLY HALF THE MOVE, AND THE ORIGINAL SENSE WOULD HAVE TURNED INTO A HALF-FINISHED PIECE。
CORE PAIN POINT: THE ESSENCE OF THE VIDEO IS "MOTION ARITHMETIC." YOU NEED TO UPGRADE THE HINT FROM A "SPEECH DESCRIPTION" TO A "SPEECH SCRIPT."。
[Practice Case Presentation: Snow Message]
Initial state: a medium view, girls crouching in the snow, with a lot of letters coming up。
Midjourney v. 8.1 generation, prompt:
A Chinese girl crouching in the pure white snow, confronted in front of her dozens of white cities and better princes,
[Turnful Case]
The wrong hint: "There was a storm, and dozens of letters on the ground were rolled off to the sky. The girl stood up and ran forward, while the drone camera was at 360 degrees fast
IT'S NOT JUST AN EXPRESSION OF TENSION IN FRONT OF THE PRESENT AI, IT'S "UNUSUAL." BECAUSE THERE'S NO SEQUENCE, AI WILL MAKE ALL THE MOVEMENTS HAPPEN IN THE SAME SECOND, SO THE LETTERS WILL BE SO WEAK, THE GIRLS WILL BE STANDING UP LIKE AN INSTANT, AND THE CAMERA TURNS, AND THE RHYTHM OF THE MOVIE WILL BE COMPLETELY LOST。
[Face Case] (advanced technique: split elements + 8 seconds time line control)
The correct approach is to define specific movements in a time-line format, which completely divides the environment, the subject and the lens。
Correct hint:
[core parameters] Environment: Screech, cold-temperature; lens: Fixed position (Static camera)。
[0-2 seconds] Acoustic mating: girls stay down, look down at the letters, red scarf moves slightly with the breeze。
[2-5 seconds] Physical interaction: The strong wind suddenly strikes, and the mail stress on the snow produces real physical rolls and shifts。
[5-8 seconds] Emotional outbreak: girls feel wind, slowly and naturally look up to far away, and red scarfs and hairs fly with the wind。
Why is the Timeline Method a downfall
Equivalent projection:
By adopting subparagraphs, we cut into three missions for eight seconds. 0-2s calculus micro-move, 2-5s full of letters, 5-8s full of characters. Every second, AI has only one core target, and the action is solid and accurate。
2. Establishment of causal logic:
IT'S THE WIND, SO IT'S THE SPIRIT, SO THE GIRL'S UP. THIS LOGICAL CHAIN IS THE KEY TO GETTING AI VIDEO OUT OF THE PPT SENSE AND GENERATING NARRATIVE TENSION。
3. Steady spatial sense:
Now that the internal movement is complicated enough, we'll lock the camera. Drop the meaningless circle and press every degree of computation into the details of the paper flipping。
Core Summary:
DON'T THROW AN A.I., GIVE IT A SCRIPT WITH A TIMELINE. TAKE THE PERSON, THE BACKGROUND AND THE CAMERA APART AND SPECIFY WHO'S THE LEAD PERSON FOR EVERY SECOND. THIS “SEPARATE EXERCISE” CONTROL IS THE WATERSHED FROM THE “TICK CARD PLAYER” TO “AI DIRECTOR”。
Summarize:
Respect for the “absolute hegemony” of the initial frame
In the Tusheng video mode, the first frame has determined all the visual tone。
2. Establishment of a single visual dominant logic
The lens must be clearly located — either “emotionally dominated” or “advertisingly dominated”。
Introduction of the time axis of the Sniper Script
THE ESSENCE OF VIDEO AI IS "MOTION-CALCULATIONS."。