Editor's note
The Kling 4.0 is officially coming online on the 15th!
Kling has always been the first choice of many creators to make "Win Drama".The play is also quite prominent.
At this stage, overseas netizens can't wait to get back on their feet.
The cavalry paved the slopes, the giant stones shouted at the city buildings, and arrows rained in the canyon. The scene was large enough, and the cameras could clearly interchange the offensive and defensive relations, causing a feeling of oppression in the great military tensions.
Subtle character movements, drawing, aiming, throwing arrows, then taking the enemy's arrows backward, inducing and hitting form a coherent response.
There is a sense of humor in the martial arts, and there is no shortage of hand-held personality performances. The mood of the singer changes naturally as the mood of the song changes, the details are also theater, and oral matching is more in place.
Even the Japanese lyrics have similarities, and the character's tone, movements and rhythm of voice match.
"Directors are full", "It's hard to tell whether AI is AI", "Narrative and character consistency are becoming increasingly playable".
Even professional directors have come to speak. Recently, at the Shenzhen Film Festival Film Market, Shenzhen's Spring Director Golden Liu shared his experience of creating the AI new film, Worm Train Liu.
The passenger and the giant insect were crammed into the same carriage, his body constantly colliding with this core lens, which he had not been satisfied with until he tested the Kling 4.0, a crowded and surprising car show that finally had the effect he wanted.
It was also the first time that a machine's heart was qualified for intuition, a racing test. The biggest feeling was that the Kling 4.0 martial arts game was more organized, allowed for more complex movements to be unleashed, the coordinates of the camera were up to date, the consistency of the characters and the scenes was quite good, and the more complex narratives and large scenes were not even mentioned.
He's a big shot, he's a big shot, he's a big shot and he's a professional.
The Kling 4.0 has a high intensity.
The city runs cool, 15 seconds of high-energy action, almost without giving anyone a chance to breathe. High-speed running, rolling, crossing obstacles, full of characters, the footage follows a crazy path.
The most intuitive feeling is that when people are free to move, they are also stable.
During the continuous flips and jumps, the hairstyle, clothing and overall shape of the person remained good, with no apparent change of person, change of clothes or disfigurement of limbs.
It is also worth paying attention to the adjustment of the footage. From the back closely followed, side to side followed, to the height bent, the machine changed several rounds, and the motion logic is basically not "off line".
In particular, the series of movements of rolling up, rolling over the levers, landing sprints, the footage follows along, and the spatial relationship between the characters and the obstacles is also clearer.
The action drama has always been one of the hardest bits of AI video.
Role-playing characters, weapon variations, offensive tactics... I've always wanted to organize an AI-like battle, often in action pieces called "Great Chaos".
But this part of the street is quite a struggle.
The men saw the pull-off, the glue, the attachment control, the turn and the disconnect unfolded, and the camera understood the close-up of the characters continuously with the shooting, with the rainbow street view and the reflection of the wetlands, the taste of the port action film.
Of course, there is no need to be a person on the big stage. Thousands of birds gather in the sky, fold up, fold up, relax and change their shape.
Such large-scale bird wave dynamics can also maintain a relatively consistent visual effect.
The official work is a blank card roll, with one piece in hand, which is a large dynamic pressure test.
For example, the high-speed chase, continuous flickering, and simultaneous swinging of giant touches into a single image, multi-subjective movements remain coherent and spatial relationships remain stable in complex scenarios.
The real gap in perception is the lens. At high speeds, with tapping, sticking over, suddenly pulling up the gaze, and then quickly cutting back to the center of motion, the movement of the lens is large, but the image is not clearly scattered.
The more intense the mirror is, the greater the impact, the more it looks like you're in a fantasy adventure while riding a mountain bike.
Overall, the Kling 4.0 is remarkable for its performance in heavy motion and complex lenses.
It's all in your head.
Don't make it hard for you to give suggestions, play them directly to AI.
After seeing the force values, a more realistic question arose: how could ordinary people take so complex action and footage so easily?
After all, for the creators of ads and videos, it's time to "listen to the commands".
What people grow up like, how they move, how the lens moves, whether the product details can be preserved... these things are too difficult to describe in terms of hints.
Especially complex movements, writing a few hundred words of hints, and sometimes it's better not to watch videos directly to AI.
So Kling 4.0 brought a huge "all-encompassing reference" -- simply put, images, videos, and character subjects can be brought to AI for "patterning" or combined use.
The images alternate character appearances, products and visual styles; video demonstration movements, cameras and narrative rhythms; subject references further specify who plays.
A single task supports a maximum of 15 reference materials, including a maximum of 10 images, a maximum of 5 videos, a maximum of 30 seconds in total, and a maximum of 7 subjects, 3 video subjects, and audio and other reference information.
It's very powerful and actually very useful. For example, video references have helped us a lot.
The video of the mountain skating speed drop is a reference to a real-life footage.
We found that in the video, some of the details that are hard to explain in writing, the model actually gets to the bottom when it falls, when it gets up, how the body bends, how the focus is shifted, and how the camera follows the beat, how the distance is adjusted as the channel passes.
Creators don't have to describe every action, every shot from scratch and give the finished image directly to AI to "play a game" to get the demo done quickly.
If only certain elements in the video are needed, it can be indicated separately.
How to carry mirrors.
These skills are especially useful for advertising and video production.
For example, an advertising company has a mature commercial in its hands, and customers like its ideas and mirrors, but want to exchange them for their own products.
Advertising companies delivered the original video and new product material directly to Kling 4.0 to quickly generate a creative demo.
Video references are very practical, but not necessarily necessary every time.For example, if you want four band members to perform together, uploading foursome material, interacting with suggestions and exchanging performance content, you can quickly verify the ideas of more than one person performing.
That's right, Kling 4.0 can also be used for "movie editing", i.e. video editing.
A newly generated video is already good, with a partial detail problem, without having to re-generate the entire video, and requests for modifications to specific elements are made, so that the model retains the rest of the content as much as possible.
We tried to add a shade to a video of a singer performing. As a result, the shade was successfully inserted into the mirror, and the rest of the content remained stable.
30 seconds of breathing generates a little more space for the story.
The action can be played, the reference can be controlled, and then there's a real pain in the ass that wants to do a little bit longer AI video, often having to put together more than ten seconds of footage.
This time, Kling 4.0 extends the duration of native video production to 30 seconds! allowing for more space for continuous long frames, multi-lens switches and a complete narrative.
The length of the video changes, and the tests are upgraded with it. Whether the characters are consistent, whether the plot is connected, whether the footage progresses at the prescribed pace, are as important as the quality of the single footage.
With the help of keyword puzzles, we can control the narrative more specifically. For example, keep the subject highly consistent in a long narrative.
The Flash version is wrong, the 4K version is fine.
Of course, in commercial creations, it is not possible for every video to be of the highest quality.
Think of the e-commerce businessmen, short video teams and advertising companies who update dozens of materials every day. One idea is five editions, ten editions, a change of character, a change of text, a change of scene, which edition works, and then continue to refine which edition.
At this point, the speed and cost of production became the real productivity. So, Kling 4.0 also brought a version of Kling 4.0 Flash, which was designed specifically for high-frequency creation.
All-in-one references can be used to generate 3×20-second videos. More attention is paid to speed and cost efficiency.
When testing, we run a demo version of Flash to see if the layout, movements and cameras meet expectations, and after confirming the direction, we cut to the full version.
Some action plays and dramatic dynamic scenes, the first round of draw cards can get good results.
In addition to efficiency, professional creators have not been able to get around the problem of quality.
Especially in the beauty, fashion and commercial advertising industries, the final videos often need to be viewed in greater detail.
The official Kling 4.0 output specifications cover 720p, 1080p, and up to 4K resolutions; 1080p and 4K will also support 10-bit HDR, with related features continuously available.
The Tahitian giant waves we generated, which are as detailed as flying saucers and the layered changes in the color of the giant waves in the sunlight, were treated very carefully and realistically.
In addition, the new model supports 21:9 film width, as well as multilingual voice generation, with different accents and dialects.
From rapid trial and error to high-quality delivery, the chain of creative tools is being further refined.
From drawing cards repeatedly, to embracing creativity
One of the most noteworthy changes of the Kling 4.0 era, we think, can be summed up in two words: controllable.
Friends who have used AI videos know that the most devastating thing is that the output is often very close to what is expected, but there are always two details that break the chain.
The Kling 4.0 update is targeting these pain points.
Large movements are more stable, all-encompassing references make it easier to express complex ideas, 30 seconds straight out provides more narrative space, and the Flash version reduces the burden of previous trial and error.
Of course, being able to accurately shoot a film every time is a challenge.
But at least from this round of testing, AI video is becoming more "human hearing" and more capable of fulfilling the specific requirements of its creators.
A little bit of "rescue material", a little bit more designing footage, organizing performances, polishing stories. This is the most anticipated change in Kling 4.0.
After all, the AI of this "camera" is so great, it is up to the creator to decide what to shoot and how to shoot it.
© THE