WHEN A LOT OF PEOPLE ARE DOING THE ROLE OF AI, THE FIRST REACTION IS TO FIND A GOOD REFERENCE, AND THEN MAKE IT HAPPEN. THIS APPROACH APPEARS TO BE FAST, BUT THE PROBLEM IS ALSO CLEAR: THE ROLE IS SO EASY TO BE LIKE OTHERS, WITHOUT ITS OWN VISIBILITY, AND IT IS DIFFICULT TO STABILIZE WHEN IT COMES TO CONSISTENCY. MORE IMPORTANTLY, THE RISK OF COPYRIGHT AND PORTRAITS MAY ALSO ARISE IF REFERENCE MAPS ARE DRAWN FROM REAL PEOPLE, VIDEO CHARACTERS OR WEB-RED PHOTOGRAPHS. THE REALLY APPROPRIATE METHOD FOR LONG-TERM USE IS NOT "REPLICATE A FACE" BUT "DESIGN A FACE". THE CORE OF AI'S SQUEEZING FACE IS NOT TO MAKE THE FIVE OFFICIALS MORE BEAUTIFUL, BUT RATHER TO MAKE THE CHARACTERS' FACES STRUCTURAL, SMUG, MEMORABLE AND CAPABLE OF...

A LOT OF PEOPLE DO AI VIDEOS WITH THEIR FULL FOCUS ON CHARACTERS, COSTUMES, SCENES AND STYLES. IT'S A LONG TEXT AND A FINE PART, BUT IT'S STILL LIKE "GENERATING A PICTURE" WHEN IT COMES OUT, AND IT'S LACKING THE QUALITY OF THE REAL IMAGE. OFTEN THE PROBLEM IS NOT THAT THE PERSON IS NOT GOOD ENOUGH OR COMPLEX, BUT THAT THE LIGHT IS NOT DESIGNED. IN THE REAL PICTURE, THE LIGHT IS NOT JUST LIGHTING PEOPLE, IT DETERMINES THE RELATIONSHIP BETWEEN THE SUBJECT AND THE BACKGROUND, WHETHER THE PERSON IS EMOTIONAL, WHETHER THE PICTURE IS HIERARCHICAL OR WHETHER THE AUDIENCE BELIEVES IT EXISTS. SO THIS SECTION, WE DON'T SPEAK COMPLEX STYLE WORDS, NOR...

THE HARDEST PART IS TO PRODUCE AN AI FIGHT VIDEO, NOT TO COPY IT INTO A VIDEO TOOL, BUT TO GET A REALLY USEFUL ONE. IF ONLY AI SAID “WRITING AN IRRITATING PUNCHWORD”, IT DID NOT KNOW WHAT THE TWO ROLES LOOKED LIKE, WHAT WEAPONS WERE USED, WHERE TO FIGHT, OR WHAT ACTION, EFFECTS AND LENS SHOULD REFER TO. THE FINAL CONTENT MAY BE POPULAR BUT CANNOT MATCH THE ROLES AND MATERIALS IN HAND. A MORE STABLE APPROACH WOULD BE TO GIVE ALL CHARACTERS, SCENES, ACTION MOVES, MIRRORS AND SPECIAL EFFECTS REFERENCES TO AI IN SUPPORT OF THE MAP, FOLLOWED BY A CLEARLY NUMBERED DIRECTIVE TO MAKE IT...

THE PICTURE IS ALREADY IN THE HEAD: ONE PERSON SITS IN A CHAIR AND THE OTHER IS STANDING NEXT TO HIM; OR A MAN IN A COSTUME HOLDS A LONG GUN AND THE GUN IS POINTED RIGHT AT THE CAMERA. YOU CAN REALLY OPEN AN AI TOOL WITHOUT KNOWING HOW TO WRITE THIS IMAGE AS A HINT. WHERE ARE THE PEOPLE STANDING? WHERE ARE YOU HEADED? HOW HIGH ARE YOUR HANDS? HOW CLOSE IS THE GUN TO THE CAMERA? IF THEY ALL RELY ON A DESCRIPTION OF THE TEXT, THE LONGER THEY ARE WRITTEN, THE EASIER IT IS FOR AI TO UNDERSTAND. IN FACT, WHEN FACED WITH SUCH PROBLEMS, WE DO NOT HAVE TO CONTINUE TO STUDY “HOW TO WRITE THIS”. THERE'S A BETTER WAY TO GET WHITE IN YOUR HEAD...

For the first time, many people saw Codex and understood it as "software to help programmers write code." For ordinary creators, however, it is more worthwhile to learn how to deal with tasks: read information, understand objectives, call tools, implement steps, check results and then revise them on the basis of feedback. After this section, you should be able to organize a vague idea into a missionable list and do a different set of short plays, video clippings, small games, etc. in the same way. To understand how Codex works, why the same sentence, "Do me a short play," might give you just one piece of a regular chat...

A LOT OF AI VIDEOS LOOK AT EACH IMAGE ON THEIR OWN, AND IT'S LIKE A FAST-TO-TURN SET OF IMAGES: PEOPLE TALK, EMOTIONS ARE STRESSED, CAMERAS ARE NOT STOPPED, AND VIEWERS DON'T HAVE TIME TO ENTER THE STORY. THE MORE MATURE WAY TO DEAL WITH IT IS TO PREVENT THE IMAGE FROM REMAINING ON HER FACE WHILE THE RED WOMAN IS WAITING. INSTEAD, IT CUTS TO THE ROOF OF THE RAINY NIGHT, SHAKES THE LANTERN, TABLE GLASSES, EMBROIDERED SHOES WITH PETALS AND FALLING RED SILK. THESE IMAGES, WHICH DON'T SEEM TO HAVE A LEAD, ARE BLANK MIRRORS OR INSERTIONS. EMPTY MIRRORS ARE NOT USED TO FILL LONG DECORATIONS, BUT TO SPEAK FOR PEOPLE. IT CAN GIVE TIME AND PLACE, AND IT CAN GIVE A LINE...

MANY PEOPLE SPEND A LOT OF TIME ON CHARACTER APPEARANCES, COSTUMES, SCENERY, AND VISUAL CLARITY IN THE PRODUCTION OF THE SHORT PLAY. LOOKING AT EACH PICTURE ALONE, THE PERSON IS FINE, THE IMAGE IS COMPLETE, AND THE LIGHT IS GOOD; BUT ONCE THE PICTURE IS GENERATED, THE CHARACTER IS PUSHED AROUND, THE FACE, THE CLOTHES AND THE BODY DON'T FIT, THE OPPOSITE IS MORE LIKE TWO PEOPLE STANDING ON THE LINE. THE PROBLEM IS NOT THAT THE MODEL IS NOT STRONG ENOUGH, BUT RATHER THAT THE HINT ONLY TELLS AI “WHAT TO DO” AND DOES NOT EXPLAIN HOW THE ACTION OCCURS. THE LIFE OF THE AI SHORT PLAY COMES FROM CONTINUOUS, SPECIFIC AND INTERRELATED CHANGES. IT'S NOT JUST ABOUT MOVING ON, IT'S ABOUT PUTTING THE WEIGHT ON THE FRONT.

A LOT OF PEOPLE DO THE SAME THING WHEN THEY DO THE AI VIDEO: IT'S TOO SHORT A SINGLE TIME, AND IT'S OVER AS SOON AS IT'S IN THE STATE. SO THE MOST COMMON WAY TO DO THIS IS TO CUT OUT THE LAST FRAME OF THE LAST VIDEO AND CONTINUE TO PRODUCE IT AS THE FIRST FRAME OF THE NEXT VIDEO. THIS APPROACH SEEMS REASONABLE, BUT WHEN IT IS REALLY CUT TOGETHER, THERE IS OFTEN A PROBLEM WITH CARTONS, PAUSES AND SPLITS. THE REASON IS SIMPLE: A PICTURE CAN ONLY RECORD A MOMENT OF SILENCE, BUT NOT A PERSON'S STATE OF MOVEMENT. AT THE END OF THE PREVIOUS VIDEO, THE PERSON MAY BE TURNING HIS HEAD, WALKING, RAISING HIS HAND, BUT THE NEXT VIDEO IS HEAVY FROM A STATIC MAP...