AIxDESIGN Bookmark Library

Write5Get unlimited access to the best of Medium for less than $1/week.Become a memberBecome a memberWhat are good AI Films in 2025?From disrupting Hollywood to contextualizing media.Fabian Mosele·Follow9 min read·18 hours agoListenShareMore2024 was the year of generative videos. From the announcement of Sora in February, to the first hands-on competitors like Luma’s Dream Machine, KLING and Gen-3 Alpha, to a second wave during the fall with Hailuo and Pika 2.0, plus the advent of open source models like Mochi, CogVideoX and Hunyuan.Big stars like Kanye, Snoop Dogg, Guns’N’Roses and Pink Floyd riding the wave with generated music videos, to advertisements like Fiverr, eToro, TED and even the Minions and the Olympics.A year of generative videos until Sora, the most anticipated model was released in December with underwhelming responses. During its unreleased 11 months, it hyper-curated whatever creatives could share, advertising it deceptively and culminating with some artist who momentarily leaked Sora on Huggingface by calling themselves OpenAI’s PR Puppets.Today we can generate videos of highly realistic scenes, something unbelievable in 2023. But Sora’s announcement, and what has come since, has been creating a false narrative that these tools couldn’t live up to: disrupting Hollywood. Many big names, naively or not, have been taking massive decision based on that. James Cameron becoming part of the board of Stable Diffusion, Tyler Perry putting his studio expansion funds on hold for fear of AI, companies firing staff believing they can replace them with AI… Meanwhile in reality big names like Toys’R’us and Coca Cola made shallow advertisements that no one liked. The web is flooded by a ridiculous amount of tutorials and spec ads on how to make films with AI instead of actually creating films with it. A media landscape stained by AI slop.So what’s in store for AI films in 2025? In this essay I will go through the biggest illusions the AI bubble has over video generative models, from its capabilities and creative empowerment, to the acknowledgment of limitations that we will not escape through “better models”. The thing is, it’s not “not there yet”. Stop using yet as if a magic spell will give these models some kind of super power. We have stunningly powerful models NOW with underlying flaws that we will always have. But this isn’t all about dropping shade on the field.On the contrary, having some well needed criticism enables us to see the beauty of generative media, the real ways in which it is changing our creative workflows and allowing us to make media that was unthinkable a few years ago.Illusion of the industry“People outside of AI just think that we click one button and the film is made. We actually generate lots and lots of videos before we curate what to put in our film…”That’s the common narrative from normie “AI Filmmakers” when accused of being uncreative. What they miss is that clicking one button, or clicking a few more buttons is pretty much the same thing. AI slop and shallow slideshows have been at the forefront of generative videos. Sorry but concatenating a bunch of generative videos don’t make you a filmmaker. Sigh.What we miss from big companies and AI evangelists preaching the idea of AI being the future of creativity, is the lack of acknowledgement of what goes into a creative process. Yes, generative tools are changing the landscape of media production in crazy ways. But it’s not as easy as “insert AI company slogan here”. Generative media models require a different approach to media production. It’s less like a camera, where you point and shoot, but more like a slot machine, where you try over and over until the machine produces something you like. Let’s focus on that a bit more, since it seems that nobody acknowledges this pretty important shift.Let’s say I want to make an ad for Coca Cola. For live-action I would need to hire many people, from camera to light and actors. Each shot needs to be planned thoroughly as it is going to be costly to re-shoot anything. If we make a 3D animated ad, as Coca Cola pioneered in the 90’s, we have more control over the set as it is all in our computers. It required some clever storytelling back than to make a good looking and heart warming story with the limited technology they had. But they managed to do so because they were good storytellers.Now if we use generative media, we also need to play with the limitations this tech has. How do we tell heartwarming stories if humans look uncanny? How do we plan to make it feel it’s all happening in one place if each clip is so different? Even if you are able to find clever solutions to these problems (something Wild Card, Secret Level and Silverside AI clearly didn’t), you are still bound to a slot machine.No matter what tool you use, it’s always about generating, seeing what comes out and trying it again until the model spits out something we can work with. Generating again and again, generating aga

Notifications

Training the Archive – Gabriel Pereira "How the image collection ImageNet re-constructs reality"