Ancestra talks a lot about the current state of AI-generated videos

AI Video & Visuals


After watching the new short film by writer/director Eliza McNitt ancestoryou can see why many Hollywood studios are interested in generating AI. Many of the shots were created and refined by the prompts alone, in collaboration with Google's Deepmind team. It is clear that Darren Aronofsky's AI-centric primitive soup production house and Google will come from normalizing this kind of creative workflow. But when you sit down to hear McNitt and Aronofsky and talk about how Shorts came together, it's not difficult to think about the possibilities of generative AI that will guide a new era of “content” that feels like they've been cooked in the lab.

Inspired by McNitt's own complicated birth story, ancestor Zoom in on the life of a pregnant mother (Audrey Corsa) as she prays to miraculously heal the heart defects of her soon-to-beborn baby. The short features many real actors who play on practical sets, but developed using Google's Gemini, Imagen and VEO models ancestorA shot of something that passes through a small dangerous hole in the mother's heart and the baby's heart. It appears in the mother's uterus blonde– Close-up of a baby with heartbeat gradually becoming part of the movie's soundtrack. And the anti-mission of women about the meaning of being a mother is visualized as a very short clip of a child born after the Big Bang, a volcanic explosion, and other women on the stars.

It's all very sentimental, but the messages conveyed about the power of motherly love are cliches, especially when they are essentially juxtaposed with a montage of computer-generated natural footage. Visually ancestor It feels like a project that is trying to prove that all the AI ​​sloping videos on the internet are actually exciting. However, the film feels like a rather weak argument in favor of a rush to reach the Slop Trough in Hollywood, as it lacks the engaging narrative substance.

When McNitt Smash cuts into quick shots of various kinds of animals in the close-up with young holes filled by microscopic organisms, you can see that these visuals explain the majority of the film's AI foundation. Each of them feels like another example of the ability of a text model to produce untextified footage that is difficult to incorporate into fully produced films. However, in a behind-the-scenes video Google shared in last week's announcement, McNitt makes more sense to her to create fakes on Google's models when she faces the difficult prospect of having to cast a real baby.

“There's no such thing as the emotions that human performances and actors evoke,” explains McNitt. “But when I wrote that I had a newborn baby, I didn't know the solution to what would happen to us. [shoot] That's because you can't make your baby act. ”

Filmmaking with toddlers brings all kinds of production challenges that are not a problem with CGI baby and doll props. But going on the Gen AI route gave McNitt the opportunity to make her film even more personal, using old photos as a newborn that serves as the basis for the fake baby's face.

A little tweaked, ancestorThe production team was able to combine Corsa with a shot of a fake baby to create a scene where both seem to interact as if they were real actors. Looking closer in the wider shot, you can see that the mother's hands appear to be floating just above the child, as the baby is not really there. However, the scene moves so fast that it doesn't stand out very quickly, so it has far less “looks” than the more fantastical shots of the film, intended to represent a hole in the baby's heart healed by the mother's will.

McNit points out how “hundreds of people” were involved in the creation process. ancestorone of the biggest take-outs of behind-the-scenes videos is that the project's production teams are relatively small compared to what they might see in more traditional short films that tell the same story. We will adopt and conceptualize more artists before crafting ancestorThe visuals of the film definitely took time to finish it over more expensive. These are extremely difficult challenges to overcome, especially for indie filmmakers and up-and-coming creatives who don't have unlimited resources at will.

GIFs displayed side by side of Google's VEO-generated AI model and video fed to the video created by the model

Image: Google

but ancestor It also feels like a case study of how generative AI stands to eliminate jobs that people would have done in the past. The argument is that in many cases AI is a tool, and jobs change rather than being exchanged. However, it is difficult to imagine studio executives truly believing in the future where today's VFX specialists, concept artists and story borders have moved to work as quick writers who have been compensated enough to maintain their livelihoods. This was a big part of attacking Hollywood films/TV actors and writers in 2023. This is why video game performers have been on strike for the majority of the past year, and they feel it is irresponsible to dismiss these concerns as they fear people simply resist innovation and change.

Aronofsky points out that cutting-edge technology has always played an integral role in the filmmaking business in video production. You will be struggling today to find modern films and series that are not produced using powerful digital tools that didn't exist decades ago. It may happen ancestorUsing generator AI that makes Google's models theoretically and with sufficient high quality training data to undoubtedly demonstrate how Google's models can do is refined enough to create footage that people actually want to see in the theatre. But the way Aronofsky becomes Stony Faced and one of Google's Deepmind researchers responds “not good” when they explain that Veo can only generate 8 seconds long clips, says a lot about where the generator AI is now. ancestor As a creative effort.

McNit feels like he's talking a bit to herself when he talks about how the output of the generative model affected the way she wrote it. ancestor. She says “both things really informed each other,” but that sounds like a very positive way to spin the fact that Veo's technical limitations need to write a dialogue that could coincide with a series of clips that are vaguely tied to the concepts of maternal and childbirth. This appears to have to strip McNitt of the intent of the core authors in favour of AI models working along what they spit out. If that's the other way around, ancestor It may have led to a much more interesting story. But there's very little about that ancestorThe story of the story, or to be honest, its visuals are so groundbreaking, so why does it feel like a Hollywood example? Should Hurry to embrace this whole technology.

Films made with more generative AI can be made cheaper and faster, but the technology that exists now doesn't seem to be able to put your ass in cinemas or create art that encourages people to sign up for another streaming service. And at the end of the day, ancestor It's an ad to hit Google's hype, and this is something that none of us should hurry up.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *