A lot of people doAI VideoWhen this is done, emphasis is placed on the hint. When the picture is unstable, the description continues; when the person changes his face, the hint continues; when the lens is wrong, the card is drawn repeatedly. The longer the last hint, the more effective it is。
IN FACT, THE KEY TO MAKING THE AAI VIDEO MORE STABLE IS NOT JUST A HINT, BUT HOW IT IS MADE。
THE TOOLS WE NEED HERE ARE AIStoryboard.
WHAT NEEDS TO BE NOTED IS THAT THE STORYBOARD HERE IS NOT MADE MANUALLY, BUT IS DIRECTLY GENERATED BY AI。
THE RIGHT PROCESS IS SIMPLE: FIRST, TO IDENTIFY THE PERSON AND ALLOW AI TO CREATE THE STORYBOARDS BY THE PERSON; THEN TO ALLOW AI TO GENERATE THE STORYBOARDS BY THE PERSON AND THE PLOT; AND FINALLY, TO GENERATE A PICTURE OF THE INTENDED EFFECTS BY THE PERSON AND THE PERSON, AND THEN TO USE THEM TO GENERATE THE VIDEO。
THE ADVANTAGE OF DOING SO IS THAT AI ALREADY KNOWS WHO THE PERSON IS, HOW THE STORY GOES, HOW THE CAMERA IS ORGANIZED, WHAT THE IMAGE SHOULD FEEL BEFORE IT IS OFFICIALLY GENERATED. VIDEOS WILL NOT BE ENTIRELY RANDOM AND WILL BE MORE STABLE IN CHARACTER, LENS AND STYLE。
CHAPTER 1: WHY IS THE VIDEO NOT LIKE AI
MANY GOOD AI VIDEOS DO NOT SEEM TO HAVE BEEN GENERATED AT RANDOM, NOT BECAUSE THEY ARE PARTICULARLY COMPLEX, BUT BECAUSE THEY WERE PLANNED BEFORE THEY WERE GENERATED。
Videos are generated only by hints, and there are three common problems。
FIRST, AI DOESN'T KNOW HOW TO SHOOT THE CAMERA。
YOU WRITE "A GIRL WALKS IN THE STREET," AND AI DOESN'T KNOW IF SHE'S GOING FROM LEFT TO RIGHT OR FROM FAR AWAY TO THE CAMERA, OR WHETHER THE CAMERA SHOULD FOLLOW, PUSH, OR STAY STILL。
SECOND, AI DOESN'T KNOW HOW PEOPLE FIT IN。
YOU WRITE A LOT OF DETAILS ABOUT PEOPLE, BUT IN DIFFERENT SHOTS, AI CAN STILL CHANGE FACE, HAIR, DRESS AND LOOK. ESPECIALLY WITH MULTIPLE-SCENARIO VIDEOS, PEOPLE CAN EASILY BE ONE BY ONE。
It's not the same
THIRDLY, AI DOESN'T KNOW HOW THE IMAGE STYLES ARE。
In the same video, perhaps the first was natural light in the morning, the second became a strong advertising light and the third was like a film poster. If the picture is not uniform, the video will appear incoherent。
So, of course, the hint is important, but it cannot take on all the tasks。
THE MORE STABLE METHOD IS TO UNTANGLE THE TASK: THE PERSON IS CONTROLLED BY THE CHARACTER'S IDENTITY BOARD; THE SCENE IS MADE BY MR. AI; THE CAMERA IS CONTROLLED BY THE AI STORYBOARD; THE STYLE IS CONTROLLED BY THE INTENDED EFFECT MAP; AND THE FINAL VIDEO MESSAGE IS INTEGRATED ONLY。
This is clearer and more likely to produce stable results than simply writing a large hint。
CHAPTER TWO: FIRST IDENTIFY THE PERSON, THEN LET AI MAKE THE STORY
1.1 Character Generation
THE FIRST STEP IN MAKING AN AI STORYBOARD IS NOT TO CREATE A STORYBOARD DIRECTLY, BUT TO IDENTIFY THE PERSON。
Because the storyboards are built around characters and dramas. The character is unknown, the story is general; the story sheet is confused。
the first thing to do is create a third view of the person, which i created with the skill of the last class。
The heroine:
prompt (image2.0 generation):
please generate an image of a 9:16 real classic movie. the characters are women, 18 to 22 years old, rich family gold, in the park scene, and the overall environment is of an ancient, windy, horticultural nature, wearing a light-blue, cold-sophisticated costume. the person's face refers by default to the facial style commonly used on the planet superi: the real nature of the face of an eastern woman, the fine but not over-redged face of a goose or egg, the natural proportion of five officials, clean skins, but not greasy, non-plastics, non-glass skins, clear eyes with emotions, make-up with restraint, the natural relationship under the lip, clean cheeks, cold air, gentle, classical, storytale。
The image is gentle and the character is going through an unspoken story. Using a semi-nature map, shallow depths, natural light, real ancient scenery, the background can be seen in tree shadows, lumbers, graphs, rails or pavilions in ancient windland parks, creating a natural and spatial environment. Clothes, embroidery, hairdressing, hair and fabric are natural, the person is soft and restrained, and the eye is emotional but not overplayed。
THE WHOLE EFFECT IS LIKE THE SCENERY OF A REAL SHOW, NOT A REGULAR ONE, A ROLE PICTURE, AN AI MODEL OR A CINEMA. THE VIDEO FOCUSES ON REAL PEOPLE'S FACES, SELF-RESTRAINT EYES, STATE-STYLE CEREMONIALS, FILM-FILMS, CLASSICAL DRAMA SCENES, AND A FOLLOW-UP VIDEO。
[Photo negative hint]
The greasy face, plastic face, fake face, net red face, overwhelming skin, glass skin, wax sense, five deformations, empty eyes, stiff expression, cheap old costumes, cinema wind, cosplay sense, costume plastic sense, head deformation, background disorder, AI sense, over-heavy, over-pretification, over-pretice, strong relativation of skin, oil over lips, melted face, mistook fingers, multiple fingers, text, water marks, logo。
Lead man:
prompt (image2.0 generation):
An old-fashioned male with a warm, cold-faced son standing under the porch, wearing a moon-white robe with a light gold-coloured and thin embroidery, a jade crown cap, a straight hair, a clean forehead, a white and jade beaut on his waist, a folded fan, a wide-faced face, five officials clean-faced, warm and a little alienation
THE IMAGE IS A LOW-SATURATED, WARM, WHITE OLD-COLOURED COLOR, WITH DARK-COLOURED WOODEN LONG PORCHES AND VANQUISHED COURTYARDS WITH WARM YELLOW LANTERN DOTTED, SOFT AND LIGHTED ON THE PERSON'S FACE, FULL COLOUR, RESTRAINT, ELEGANCE, SHALLOW VIEW, DEFUNCT BACKGROUND, SOFT-HOTNESS, FILM-SENSION, HIGH-CLASS ANCIENT WIND. COMMERCIAL PHOTOGRAPHIC SENSES, HIGH-CLASS MAKE-UP PHOTOGRAPHS, REALITY, FILM SENSES, SKIN TEXTURES, HAIRLINE CLARITY, FABRIC AND EMBROIDERY DETAILS, SINGLE PERSON, CLEAN DRAWINGS. NEGATIVE: BINARY, CARTOON, FALSE-FACED AI, PLASTIC SKIN, GREASY, OVER-REFINING, STAR-FACED, RED-FACED, SMALL-HEADED, SHORT-LEGGED, WIGGED, HOODED, MUTATED, TACKY-PAINTED, MULTIPLE-FACED。
character three-view hint:
The three-view image of the person, the white-story map, the left-hand part of the figure shows the fine figure of the human body. Its facial features, skeletal structure, dementia, hair and head dressing must be strictly consistent with the nature of the skin texture, hair holes and hairline details. On the right side, the full-body picture of the role is presented in a standard A-pose position, organized in three precise views: positive view, side view and back map. Supermodel body size, complete clothing design, fabric level, shoes and overall dress style must be strictly followed. All clothing and accessories present a real sense of physical material quality。
1.2 Story Generation
ONCE YOU'VE GOT THE CHARACTER'S IDENTITY CARD, LET AI CREATE A STORY BASED ON THE CHARACTER。
The story doesn't need to be complicated, the clearer the better. In particular, short video, it's best to break it down to six or eight shots, each of which can be shown in a picture。
This is the way to warn AI (to throw a person with a hint to the big linguistic model ChatGPT/gemini):
The plot produces a hint:
PLEASE GENERATE A SHORT STORY FOR THE AI VIDEO BASED ON THE FOLLOWING PROFILE. IT'S SIMPLE, CONSISTENT, GRAPHIC, SUITABLE TO SPLIT SIX TO EIGHT SHOTS. DO NOT DESIGN OVERLY COMPLEX SCENARIOS, FOCUSING ON THE BEHAVIOUR OF THE PERSON, EMOTIONAL CHANGE, SPATIAL RELATIONS AND GRAPHIC ATMOSPHERE。
Once the story is produced, check if it's suitable for video。
A good story is not a good text, but a good picture。
Therefore, the focus of this step is on:
First you fix people, then you create scenes that can be depicted。
CHAPTER III: GENERATING THE AI STORYBOARD BY CHARACTER AND BY PLOT
ONCE IT'S SETTLED, WE CAN GET AI TO CREATE THE STORYBOARD。
THE STORYBOARD HERE IS NOT HAND-DRAWING, BUT A MULTI-GAUGE SPECTROSCOPY BASED ON THE CHARACTERS AND THE PLOT. ITS ROLE IS NOT TO GENERATE FINAL FILM, BUT TO PLAN THE LENS AHEAD。
A GOOD AI STORYBOARD THAT SHOULD INCLUDE THESE:
The position of the person in the picture
The view of each shot
(a) Personal movements
Camera exercise
Image focus
Simple text labels。
You can use the following:
Storyboard generation of hints (GPT image 2.0, continued with the dialogue in the previous chapter):
PLEASE GENERATE AN AI VIDEO STORYBOARD, BASED ON THE ABOVE-MENTIONED PROFILE AND PLOT, THAT REQUIRES HAND-DRAWN STYLE. STORYBOARDS ARE SIX TO EIGHT SPECTROGRAPHS, EACH OF WHICH CLEARLY SHOWS THE LOCATION OF THE PERSON, THE VIEW, THE MOVEMENT, THE MOTION OF THE LENS AND THE FOCUS OF THE IMAGE. THE IMAGE USES A CLEAR SPECTROGRAPH STYLE AND DOES NOT MAKE A FINE POSTER. AT THE BOTTOM OF EACH GRID, ADD A SHORT DESCRIPTION OF THE CHINESE LANGUAGE, INCLUDING THE SCENERY, THE MOVEMENT OF THE PERSON AND THE MOTION OF THE LENS. AN ARROW CAN BE USED TO MARK THE MOVEMENT OF PEOPLE, THE MOVEMENT OF LENSES AND THE MOVEMENT OF THE ENVIRONMENT。
When the storyboard is created, let's not just see if it looks good, but see if it's clear。
Focus on three sites。
First, see if the sequence is consistent。
The first one is in the room, the next one suddenly appears on the street and there is no transition between them, and this is not a good story sheet。
Secondly, see if the person's movement is clear。
For example, walking, turning back, raising hands, sitting down, stopping, these moves have to be visible. The clearer the move, the more stable the video is。
Third, see if it makes sense。
For example, push, pull, move, follow, and circle are consistent with the image. Don't use complex lenses in every cell, otherwise the video can be messy。
Storyboards are primarily for the camera, not for the final quality. The character details are given to the identity board and the image style is given to the intended effect. It's enough for the storyboard to be clear。
Chapter 4: Generation of scenarios with storyboards and characters
Don't rush video after you have the storyboard. The next step is to generate a projected impact map。
The intended effect is to harmonize the image style of the whole video。
Storyboards deal with "How to shoot."。
The character ID card deals with "who's in the picture."。
The desired effect map addresses “what the final picture should feel”。
For example, if you want a real movie sense, the expected impact map is a natural light, a real colour, a clean image and a film quality。
You want a sense of life in the morning, you want to see a picture of the effect of the light, the quiet space and the warm tone。
If you want to feel the streets at night, the expected effects are to reflect the light, the dark level, the reflection of the ground and the warmer。
THAT'S HOW IT'S SUPPOSED TO BE
prognosed effects diagrams generate hints (generated using nanobanan pro)
PLEASE GENERATE AN AI VIDEO IMAGE OF THE INTENDED EFFECTS BASED ON THE FOLLOWING PROFILE AND STORYBOARD CONTENTS. IMAGES REFLECT THE OVERALL STYLE, LIGHT, TONE, MATERIAL SENSE AND SCENERY OF THE FINAL VIDEO. THE CHARACTERS MAINTAIN THEIR IDENTITY ON THE IDENTITY BOARDS, THE IMAGES ARE NATURAL, THEY HAVE A SENSE OF FILM, THEY ARE NOT EXAGGERATED AND THEY ARE SUITABLE FOR THE STYLE OF SUBSEQUENT VIDEO PRODUCTION。
THE EXPECTED EFFECT MAP DOES NOT NEED TO INCLUDE ALL THE LENSES. IT'S MORE LIKE A STYLE SAMPLE, TELLING AI WHAT IT'S SUPPOSED TO BE。
Chapter 5: When the video is eventually generated, the hint is simple
When a personal identity card, a plot, a storyboard and a projected impact map are ready, no direct video is generated。
HERE'S ANOTHER STEP: LET AI CONTINUE TO DESIGN AN APPROPRIATE DIALOGUE SCENARIO BASED ON THE STORYBOARD。
The reason is simple: the storyboard is primarily about images and lenses, and it usually does not produce a complete description of the lines. Even if there is some text on the storyboard, such as Seedance 2.0, video tools may not be accurate in identifying the text in the picture. If you rely directly on the text in the storyboard, there is a risk of misinterpretation when the video is generated。
So the more stable approach is:
STORYBOARDS ARE FOR IMAGES, AI SUPPLEMENTS THE DIALOGUE, WHICH THEN JOINS THE VIDEO AS A REMINDER。
DIALOGS GENERATE HINTS (CONTINUE DIALOGUE WITH THE AI LARGE MODEL):
PLEASE DESIGN A DIALOGUE SCENARIO FOR VIDEO IMAGES BASED ON THE ABOVE-MENTIONED PERSONAL IDENTIFICATION BOARDS, DRAMAS AND STORYBOARD CONTENT, WHICH NEEDS TO BE GENERATED USING A LENS. DIALOGUE IS NATURAL, SIMPLE AND CONSISTENT WITH THE IDENTITY OF THE PERSON AND THE CURRENT CONTEXT, AND IS USED TO ASSIST AI IN UNDERSTANDING THE EMOTIONAL AND DRAMATIC ADVANCEMENT OF THE PERSON. DO NOT CREATE SUBTITLES, DO NOT LET ANY TEXT, TITLE, SUBTITLE OR WHITE BUBBLE APPEAR IN THE IMAGE. THE FINAL VIDEO WILL ONLY HAVE TO SHOW THE PERSON'S STATE OF CONVERSATION AND EMOTIONAL CHANGE AND NOT GENERATE TEXT IN THE IMAGE。
Once the dialogue is generated, it is placed with the character's identity board, the storyboard, and the intended effect map in the video-generated hint。
Video generation of hints:
The video is based on the personal identity card (Figure 1) for women and (Figure 2) for men, storyboard (Figure 4), projected effects (Figure 3) and dialogue. The identity of the same person remains unchanged and is generated by the sequence of the lens, the movement of the person and the motion of the lens in the storyboard. Image style reference pre-intensity maps to maintain a unified light, a unified tone and a real film sense. The context of the dialogue is used only as a reference to the person ' s expression, mouth, emotional and interactive rhythm, and does not produce any text, subtitles, titles, white bubbles or text in the image. The movement is natural, the lens is stable, the picture is consistent and does not change the person, dress, scene and style at will。
Chapter 6: Common issues and complete templates
THERE ARE THREE MOST COMMON PROBLEMS WHEN USING THE AI STORYBOARD。
THE FIRST QUESTION IS TO GET AI TO GENERATE THE STORYBOARD DIRECTLY FROM THE START。
THE STORYBOARDS THUS GENERATED ARE OFTEN EMPTY BECAUSE AI DOES NOT KNOW WHO THE PERSON IS OR WHAT THE STORY IS. THE RIGHT THING TO DO IS TO GIVE A CHARACTER FIRST, THEN ALLOW AI TO CREATE A STORYBOARD, AND FINALLY TO CREATE A STORYBOARD BASED ON THE PERSON AND THE PLOT。
The second problem is that the storyboard is too complicated。
STORYBOARDS ARE NOT END-EFFECT MAPS, BUT ONLY CAMERAS AND ACTIONS. THE DETAILS OF THE PERSON SHOULD NOT BE PLACED ON THE STORYBOARD, OTHERWISE AI MIGHT HAVE USED EACH OF THE CELLS AS A DIFFERENT REFERENCE, LEADING TO INCONSISTENT FIGURES。
The third issue is the conflict between the final message and the preceding one。
The characters, stories, storyboards and pre-intended impact maps had already been set before, and in the end, no more random changes had been made. The more uniform the hint, the more stable the picture。
The following is a simplified template that can be reproduced directly。
1. The third view of the person
The three-view image of the person, the white-story map, and the left-hand part of the part shows the face of the fine-skinned portrait of the person. Its facial features, skeletal structure, dementia, hair and head dressing must be strictly consistent with the nature of the skin texture, hair holes and hairline details. Real skin pores and fur. On the right side, the full-body picture of the role is presented in a standard A-pose position, organized in three precise views: positive view, side view and back map. Supermodel body size, complete clothing design, fabric level, shoes and overall dress style must be strictly followed. All clothing and accessories present a real sense of physical material quality。
2. Story generation tips
Please generate a short story for the AI video based on the following profile. It is simple, consistent, graphic and suitable to split six to eight shots (15s). Do not design overly complex scenarios, focusing on the behaviour of people, emotional changes, spatial relations and the visual atmosphere。
3. Storyboard generation of hints
PLEASE GENERATE AN AI VIDEO STORYBOARD, BASED ON THE ABOVE-MENTIONED PROFILE AND PLOT, THAT REQUIRES HAND-DRAWN STYLE. STORYBOARDS ARE SIX TO EIGHT SPECTROGRAPHS, EACH OF WHICH CLEARLY SHOWS THE LOCATION OF THE PERSON, THE VIEW, THE MOVEMENT, THE MOTION OF THE LENS AND THE FOCUS OF THE IMAGE. THE IMAGE USES A CLEAR SPECTROGRAPH STYLE AND DOES NOT MAKE A FINE POSTER. AT THE BOTTOM OF EACH GRID, ADD A SHORT DESCRIPTION OF THE CHINESE LANGUAGE, INCLUDING THE SCENERY, THE MOVEMENT OF THE PERSON AND THE MOTION OF THE LENS. AN ARROW CAN BE USED TO MARK THE MOVEMENT OF PEOPLE, THE MOVEMENT OF LENSES AND THE MOVEMENT OF THE ENVIRONMENT。
4. Generates the pre-intuitive effect maps
PLEASE GENERATE AN AI VIDEO IMAGE OF THE INTENDED EFFECTS OF THE INDIVIDUAL SPECTROSCOPY, BASED ON FIGURES 1, 2 PERSON IDENTIFICATION BOARDS AND 3 STORYBOARDS. IMAGES REFLECT THE OVERALL STYLE, LIGHT, TONE, MATERIAL SENSE AND SCENERY OF THE FINAL VIDEO. THE CHARACTERS MAINTAIN THEIR IDENTITY ON THE IDENTITY BOARDS, THE IMAGES ARE NATURAL, THEY HAVE A SENSE OF FILM, THEY ARE NOT EXAGGERATED AND THEY ARE SUITABLE FOR THE STYLE OF SUBSEQUENT VIDEO PRODUCTION。
5. Video generation alerts
Please generate a video based on the identity card of the woman and the man, the storyboard, the projected impact chart and the conversation. The identity of the same person remains unchanged and is generated by the sequence of the lens, the movement of the person and the motion of the lens in the storyboard. Image style reference pre-intensity maps to maintain a unified light, a unified tone and a real film sense. The context of the dialogue is used only as a reference to the person ' s expression, mouth, emotional and interactive rhythm, and does not produce any text, subtitles, titles, white bubbles or text in the image. The movement is natural, the lens is stable, the picture is consistent and does not change the person, dress, scene and style at will。
Conclusion
IN ORDER TO BE STABLE, THE AI VIDEO CANNOT BE BASED ON JUST ONE COMPLEX HINT。
THE BETTER WAY IS FOR AI TO IDENTIFY THE PERSON, THEN ALLOW AI TO GENERATE THE STORY SHEET, THEN THE STORYBOARD, BASED ON THE PERSON AND THE STORY SHEET, AND FINALLY THE PICTURE OF THE INTENDED EFFECT, USING THE STORYBOARD AND THE CHARACTER. IN THIS WAY, THE CHARACTERS, THE DRAMAS, THE LENSES AND THE STYLES ARE CLEARLY REFERENCED。
The place where the storyboard is really good is not to complicate the picture, but to make it clear before the video is generated. Previously, it was after the video had been produced that it was known where to turn; it was now possible to look at the storyboard and solve the problem in advance。
WHEN CHARACTER IDENTIFICATION BOARDS, AI STORYBOARDS AND PRE-INTENDED IMPACT MAPS ARE MATCHED, THE AI VIDEO IS NO LONGER JUST A LUCKY CARD, BUT MORE LIKE A PLANNED CREATION。