Can AI actors beat real stars? Hollywood actors are having a complete falling out

The issue of AI actors in Hollywood is currently making a big fuss.

A western film called “Into the Ground” used AI to “resurrection” the young Val Kilmer, and his family gave the nod of approval on the grounds that he loved to toy with new technology that could tell stories during his lifetime.

In the trailer, the digital Kilmer provocatively said, “Don’t be afraid of the dead, and don’t be afraid of me.” However, the robot version of Kilmer that the audience saw had straight eyes and a wrong voice, as if it was a robot in the wilderness of the 1920s. Many people were frightened when they saw it.

On the other side, rising AI star Tilly Norwood is about to star in the feature film “Dislocation.” When she debuted last year, Emily Blunt said it was “really scary.”

After all, what everyone fears most is that AI will one day take away the jobs of real actors.

Tim Webb of Framestore, a major British visual effects company, believes that,The AI ​​model only learns the surface of things, while the image model learns the pixels of Internet pictures. It cannot capture the delicacy of real people’s emotions, and it is far from being able to truly “act out”.

This is exactly the opposite of performing by digging into your heart to find the truth.

But we have to admit that the performances produced by AI are becoming more and more like real people.

Many companies are now obsessed with using AI to make videos, and Higgsfield AI is one of them. They recently said that they have mastered the uncanny valley effect. It’s that feeling of looking like a human being, but something is wrong, which makes people feel uneasy. Many previously detailed CG characters couldn’t escape this problem, such as the younger Indiana Jones in “Raiders of the Lost Ark”.

Higgsfield has launched two complete AI movies this year, and the head of content, Adil Alimzhanov, is very confident.

According to him, the new world of all-AI movies can relax creators and eliminate the entire production process that is slow and cumbersome in Hollywood.

But these two films are only available online.

There is a 95-minute fantasy action film “Hell and Hell” and a 115-minute Guy Ritchie-like crime comedy “The Cali Mountain Kid.”

Not counting the time spent writing the script, the former took the equivalent of two weeks from shooting to post-production, while the latter only took a month. The 23-year-old Alimzhanov also used his face as the protagonist of “Trial from Hell”.

He understands that live-action actors will be worried, but feels that there is no way to stop the general trend: we cannot resist new technologies just because they change the industry.Think about when cars first came out, if everyone resisted cars, now we would all still be riding horses.

Although “Hell Trial” and “Calli Hills Kid” are very long advertisements for Higgsfield’s own AI tools, it is also obvious that the AI ​​performance has improved rapidly.

In “Hell Trial”, which was released in May, the skateboard boy character can only perform one emotion almost every 15 seconds.

“Kalishan Kid”, released three months later, is much better. The dialogue is natural, the comedy rhythm is online, and the laughs can really make people laugh.

The three running rap protagonists in the film are based on streetball player Matt Chiatipis, MMA player Israel Adesanya, and anchor N3on.

Higgsfield is originally targeting Generation Z creators who are accustomed to short videos and have been exposed to AI since childhood.

But looking at it in the AI ​​era, the identity of “actor” itself is a bit strange.Since each film can directly generate its own characters, why bother finding real people to play them?

The script of “The Calisan Kid” was written by a real person, Tim Planagan, and the film was created using Seedance 2.5, one 30-second segment and one 30-second segment.

But it is not easy at all to use prompt words to create a natural and dramatic scene.

First of all, you have to make good character materials, take full-body shots and facial close-ups, and then feed them to the AI ​​to ensure that the character’s appearance will not change randomly. After that, these materials were used as a benchmark. Each 30-second shot had to be written with a prompt word of up to 4,000 characters, part of which was written with the help of the big model.

It clearly describes the actions, how to move the camera, how to use the lighting, and what emotions the characters should have… Repeatedly change the prompt words and regenerate them over and over again, trying to get rid of the awkward fake feeling.

This process is very much like playing with a puppet, slowly adjusting one string at a time.

Having said that, I feel that if there are AI movies everywhere in the future that no one likes and have weird looks, I really don’t know what the audience will watch.