Have you ever brushed a shivering fire AI short filmNo time to live anyway
A SINGLE VIDEO BREAKING 100 MILLION HITS, A COMPLIMENT OF 1600 W+, AND THE AUTHOR SENT ONLY 11 PIECES, AND THE POWDER BROKE MILLIONS。
THIS IS THE KIND OF IA SHORT PLAY THAT CAN GO OUT IN CIRCLES, AND IN ADDITION TO THE EXCELLENT VIDEO ITSELF, IT'S MAINLY A PLAY-EMOTIONAL PERSON。
NOW A LOT OF PEOPLE ARE TRYING TO DO AN AI SHORT FILM, BUT IT'S HARD NOT TO WRITE A SCRIPT。
ACTUALLY, FOR THE FIRST TIME, YOU DON'T HAVE TO DO THIS TOO COMPLEX AI SHORT FILM, STARTING WITH BRAINLESS AI VIDEO。
WHEN YOU GET FAMILIAR WITH THE VARIOUS AI TOOLS, YOU TRY TO WRITE SCRIPTS AND MAKE MORE COMPLEX CONTENT。
FOR EXAMPLE, IT'S BEEN PRETTY HOT LATELY, AND IT'S QUITE A LITTLE BIT OF ALL THE A.I. RECALIBRATING HUMAN ACTION。
I DIDN'T KNOW IT WAS AI WHEN I FIRST GOT IT。
The people in there danced and twisted their backs, and the thing was so bad, they threw their hands, turned around, and it was like real people。
AND THIS LITTLE DANCE VIDEO GENERATED BY AI, WHICH IS GENERALLY VERY GOOD, MAKES HUNDREDS OF THOUSANDS OF HITS AT RANDOM, ESPECIALLY EASY TO PAY。
TODAY, TEACH YOU HOW TO DO THIS AI RESET VIDEO。
For example, I moved the classic two-person bath section of Wang Zuzheng and Chang Maggie in "The White Snake" to the Sun and Limpolis of the Evangelical Warriors of the New Century。
IT'S DIFFICULT TO DO AN AI VIDEO REGRAVING IF IT'S JUST A CARD, BECAUSE THE ACTION IS REGRAVINGBut the person that was created looked and was dressed, and he saidEasyGot taken by the original video。
Recommended LibTV "Deep action capture" on the line。
It's a function that can..The depth dynamics in the reference video are presented separately,Directly generates deep video。
Remove the visual features that make you susceptible to pollution, such as the person's appearance, clothing and background, so that you can drive your role with a pure track。
And it's completely free of charge, without depleting any points。
Experience Links:
Once you enter the web page, upload the material, click the frame-by-frame, and go down to depth action to capture, and you can use it without a brain。
I don't knowI'm going to show you my reality film, and I'm going to teach you how to make it。
I. The classic clips
The first video, the classic two-person bath in "The Green Snake" by Wang Zhu Xian and Maggie, was taken from the opening。
This time, a cross-section, the movement in the original film was moved on to tomorrow's evangelicals and Lingpolis。
The double action of this scene involves a great deal of covert and spatial relationships. Two persons are prone to arm-dressing, body-linking or sudden confusion。
Moreover, the photocopy, costume and character recognition of the original film of the Green Snake are very high。
If common video reference is used, these visual features can easily be contaminated together。
So we turn on LibTV, upload the reference video. And then the lightFrame by FrameDown toDepth Action CaptureSo you can generate a deep video。
The result is very slippery。
The video nodes were then added, and in-depth videos and two profiles were added to the nodes。
The following hints are given to generate a fully controlled video of the images and mirrors。
Here's a little suggestion, and as we ran, we found that Seidance 2.5 was quite sensitive to the hot roleIt may not be possible to use the original role map directly。
You can run the role again with the graphic function and avoid the character name in the hint and write the hint by proxy。
If Seadance 2.5 is not a good audit, it can also be generated from other models in LibTV, such as Wan 3.0 and MiniMAX H3, which are very useful。
Finally, look at the results。
Intuitively, the two-person position, the body's direction, the turning back and the pace of movement in each other's proximity were largely preserved and the overall effect was good。
In the past, such footage was produced, often requiring repeats of the cards, followed by a search for marginal versions of multiple results. With in-depth video now used as intermediate control, the first generation is close to being available, followed by fine-tuning of the hints。
In this way, when the right reference footage is found, it can be deposited into reusable dynamic material. The new role was then replaced without having to re-engineer the entire logic of the movement。
Two, King Aida and Leon
The position ahead depends on whether the base is maintained. The second video, I'd like to test if I can catch the image of each shot when the deep video encounters a continuous lens。
So, I picked a bit of a more strident rivalry with Seo Chang-Jing in Fair Sword III。
This is supposed to be a white moon for a lot of people, and can we keep the same feeling after two completely different characters? I decideLet King Ada and Leon play.
There's no big move on the surface. It's actually a test model。
Actors rely mainly on the eyes and subtle expression of their emotions, and the lens is constantly changing the view of two people。
The whole thing looks good. The sight of two people is basically connected, and the camera switch is more natural. In particular, the reaction of the King of Aida is in keeping with her original mysterious and proactive character。
But Leon's face is still a little stagnating. You can try it yourself。
The method of production and the logic behind it are similar, with the original film being converted into an in-depth video, access to the personal material of the King of Aida and Lyon and the transformation of the scene into a dark bar。
A lot of details have been given to people because of the proximity。
To replace the scene, it is important that reference maps be given to the scene。
The introduction is as follows. With in-depth video reference, it is not necessary to write a particularly complex description of the action, but to identify the person to whom each picture corresponds, and to add a new set of scenes。
This case is tested, and I don't think deep video is just a way to hold people on. Many of the details are difficult to describe with a hint and can now be sent directly to the model via reference video。
DO YOU REMEMBER THAT THERE WERE A LOT OF PEOPLE ON THE INTERNET AT THE PREVIOUS TIME, AND THAT AI DIDN'T EVEN MAKE A GOOD PHONE
The tip was written for half a day, and the person still had no idea what to do with his cell phone, which often generated a picture of the thunderstorm。
Now there is a more violent solution to this problem. Find a live video of the right move, turn it into a deep video, and use it to control your role。
III. Veterans and dance card production
Finally, let's look at deep action to capture the most widely used scenario: martial arts and dance。
Both types of video are very demanding for the rhythm of action. It is easy for the model to process the whole of the action at the same speed, simply by creating a hint。
The fast place is not going to get up, it is not going to stop when it's going to stop, and it turns out to be the man who did the command mechanically。
In-depth video provides a frame-by-frame process that allowsTHE RESULT WAS MORE HEAVY, AND FINALLY AI WAS ABLE TO GET A LITTLE BIT。
This is a chorus that I created with a deep video, replacing the characters with Machima and Pava in the chainsawman。
As you can see, the real people on the left are jumping, the right saws are jumping, the movements are almost synchronized。
THE MOST UNEXPECTED THING WAS NOT THE BIG MOVES, BUT THE SMALL PLACES. WHEN YOU GET DOWN, THE WEIGHT GOES DOWN, AND THERE'S A SLIGHT TIP UP THE UPPER HALF TO FIND BALANCE, NOT THE FLAT SPEED THAT'S COMMON IN AI。
And if you block the original on the left, you probably won't realize at first that it's taken from a direct shot of the real scene。
And then I threw in the play in the hideaway, and I captured it in depth, and then I recreated it in the style of the Shaolin。
This is the scenery in the original film, and it turns directly into a swirl-style taste, with a change of make-up, a change of weapon from a sword to a machete, and a change of costume to a swirl classic。
Because the dynamics of the martial arts play are more intense than the danceI thought the model would fail, but it was solid. A sense of fighting power, rhythm, all of it。
The machete swings past arc, and when the other side moves, the body moves in the direction of force。
YOU CAN'T GET AI FROM ZERO TO DO IT RIGHT, BUT NOW THERE'S A DEEP VIDEO, EVERYTHING'S POSSIBLE。
These two videos are much simpler than the one before, and after replacing the charactersAll you have to do is talk to LibTV about how deep video tracks are generated, and the model will read itself how to move。
If you've tried to control the rhythm of dance and martial arts with a hint, you should know how painful it is. There is now a less painful solution to this problem。
IV. Smart editing
After talking about deep action capture, by the way, a new feature I painted on LibTV, Smart Clip。
It would have been a shame not to mention it after a deep video。
you know, i've been making videos on ab for a long time. every time you make a video, it's easy to make a video, and a clip is where you really eat time。
SOME OF THE FRIENDS WHO DO THE INFORMATION STREAM, THEY HAVE DOZENS OF TEXT MESSAGES A DAY, AND EACH ONE OF THEM HAS TO BE CUT MANUALLY, AND IT'S VERY MUCH TO BE HOPED THAT AI WILL BRING THIS UP。
LibTV, which is the main channel for editing, can automatically understand the content of your material with AI, and help you complete the whole set of processes from rough to fineOne key clip, and no points。
It's easy to use. Add a "video editor" node to the painting, and then choose "broadcast video" and then connect all the footage。
All you have to do next is tell it in one sentence. For example, "Help me remove the pauses, salivas and repetitions in the material, and make a full oral video."
WHEN SUBMITTED, AI AUTOMATICALLY COLLATES ALL MATERIAL, SELECTS WHAT WORKS AND FINISHES THE CLIP. AFTER PROCESSING, CLICKING ON "OPENING SMART CLIP" ALLOWS YOU TO VIEW THE FILM DIRECTLY, AND YOU CAN GO INTO THE TIME LINE AND CONTINUE TO ADJUST。
I've produced a few pieces of cosmetics, so it can cut me straight into a whole piece。
When the material is available, LibTV analyses the type of each video and separates it from B-roll。
The text by word is then extracted from the mouth, while the content of the other material is understood, and preparation is made to remove the pause and automatically match the insertion lens。
Soon, LibTV will give a hard cut to the A-roll。
It not only cuts mouths and gas, but also identifies the silence of the tail of the head by paragraph, which is logically calibrated。
When I confirmed that the A-roll broadcast was correct, it began to insert B-roll, giving a clear list of images。
Each piece of material, starting with a few seconds, is clearly marked under which line should be hung。
It will also determine the intensity of the correlation between material and content at the oral level, facilitating subsequent examination and adjustment。
It is clear that the video has developed a complete carrying logic. There are also some fancy effects。
Although the visual effects are currently weak, an initial version of the text is sufficient。
Significant savings were achieved in the cost of late editing。
V. Summary
OK, today I taught some of LibTV's little practical tricks。
AFTER THESE AI VIDEO TOOLS ARE USED, THE NEXT STEP IS TO LEARN TO WRITE ALL KINDS OF SCRIPTS AND MAKE SOME MORE DIFFICULT AI SHORT FILMS。
I HOPE THAT'S WHAT YOU WANT TO DO