Generative AI has made great and rapid progress in recent years. This also includes video generation. What does this mean for the film industry? We have worked with Jan-Hendrik Holst spoken. He is an innovation manager and producer and thus brings both experience from the film industry and a practiced view of transformation processes. In the first part of the ARIC interview, he talks about the opportunities that AI offers, especially for small film companies, the pitfalls European models and dealing with change. This conversation is the first part of a two-part interview. You'll find part two on our blog soon.
ARIC Hamburg; Who are you and what connects you to AI?
I am Jan-Hendrik Holst, innovation manager with a focus on generative AI at an international retail company. There I develop and manage strategic AI initiatives,
particularly in the area of marketing.
My background is in the film industry. After studying Business Administration, I studied Serial Producing at the Filmakademie Baden-Württemberg and then worked as a producer
, including at Studio Hamburg on the international co-production Davos 1917.
When ChatGPT came out, was there an awareness in the film industry that this could be important?
In the film industry, the initial reaction to ChatGPT was divided. I experienced a lot of interest, but also a lot of concern, especially among authors. The fear of being replaced was palpable.
For me personally, the end of 2022 was a turning point. When I first saw what these tools were already capable of, I realized that they would not only change the workflow, but potentially the business model of entire industries.
I was largely alone in this conviction at the time. There were few who thought the same. Many thought ChatGPT was a tool for texts, a toy for nerds. In the meantime, the perception has shifted significantly. It’s no longer just about generating text, but about images, videos, entire production chains – a whole transformation.
Then at some point I realized that being on track with the whole AI topic, keeping up with updates and being a producer at the same time was getting out of hand. That is very
time-consuming. That’s why I decided to focus almost entirely on AI and innovation. Because it was already clear to me that something was going to change fundamentally, whether that was just due to the AI transformation or to many other transformations that are currently taking place.
“The classic gatekeeping structures of the film industry are increasingly crumbling.”
What makes you think that?
Technological progress has already heralded a change. Social media, YouTube etc. are merging industries and blurring boundaries. The classic gatekeeping structures of the film industry are increasingly crumbling.
New talent no longer just comes from film schools, but from the creator economy. This trend is accelerating massively with generative AI. Suddenly, talented people who previously had no chance of a career in film will be able to produce high-quality visual content and, conversely, creatives from the film industry will soon be in high demand in other sectors. I think that boundaries are shifting or even completely new industries are emerging. I believe that this innovation or even disruption will create completely new structures. I am convinced that the boundaries will be much more open in the future and that completely new business models will emerge. The creator economy didn’t exist back then either – and can now be found in many industries.
And do you personally believe that AI is more of an innovation or more of a disruption for the film industry?
For the film industry, I personally believe that it is more of a disruption. This is not negative per se, but it is important to understand the difference. Innovation optimizes existing processes: faster, cheaper, more efficient. Automated color correction instead of manual work. Disruption continues: new formats are being created by AI that are changing consumer behavior. The competition for the movie is not always just the better movie, but a different way of spending time. YouTube, for example, is not replacing movies per se. But three hours of YouTube in the evening means no movie. The fragmentation of attention then becomes the real problem and AI accelerates and simplifies the creation of competing products.
But of course AI also has a big impact on the way we work. Two years ago, I was even more skeptical. In the meantime, some of the predictions made back then are actually coming true. But that doesn’t mean it’s a problem for everyone in the industry. Smaller companies in particular can benefit because they are more agile and can adapt to new workflows more quickly. Larger companies now have to make much more of an effort, even if some have already taken the first steps.
What are these predictions from two years ago that are now coming true?
If you believe the Open AIs and Googles of this world, we would already be in a completely new world order. Of course, there’s always a lot of marketing involved. That’s why I look at it with interest, but critically. But what happened with the text models at the beginning, then with images and now with the video models, is already very far-reaching. New releases, such as Sora or Veo 3.1 in February 2024, are really big leaps. The quality of these models has exploded in 18 months. Whether this is already cinematic is, of course, open to debate. But these developments are rapid and you have to face up to the pace.
Now we could also see another extreme shift. With Seedance 2.0 from ByteDance and Kling 3.0, models are now coming that could technically redefine the game once again. If the “multi-shot” capabilities and visual quality are confirmed in practical tests, then we will have to adjust the benchmarks once again and correct some statements. But first let’s see if it holds up in the complex production workflow.
Where are these changes happening in the film industry?
In the last few months, I have seen increased attention being paid to the topic in the German film industry in particular, of course in its theoretical form at film festivals, panels and the like. However, theory is now becoming practice in many areas. Numerous AI studios have been founded in the USA, some VC-financed, many bootstrapped. Major studios such as Lionsgate, Disney and Netflix are entering into strategic partnerships. Lionsgate, for example, announced a high-profile collaboration with Runway in 2024. One year later, however, it has become clear that this does not work as easily as expected. The catalog is not enough to train your own model. This is an important lesson for the industry.
More is now also happening in Germany. For example, Beta has founded the start-up Chapter 41 together with Drive Beta and has also announced that it will enter into a strategic partnership with the Baden-Württemberg Film Academy to explore AI and innovation in the audiovisual sector. Storybook Studios in Munich is one of the few companies that has had its finger on the pulse from day one and has attracted attention far beyond Germany. The step from theory to practice is happening right now. Even if the business models have not yet been finalized.
Tools such as Sora 2, Veo, Seedance, Kling etc. contribute significantly to this, but also the increasing cost pressure. Two trends are coming together here: The film industry is not spared from rising costs and falling orders. In times like these, technology that promises to cut costs has enormous appeal. This is giving momentum to AI adaptation in Germany and Europe. But not always under the best conditions. In Germany in particular, the order and funding structure means that pure cost reduction is too short-sighted. When budgets fall, funding and order volumes also fall. At some point, this erodes the business model itself.
Are video generation models already being used to generate market-ready films?
Fully AI-generated films are not yet ready for the market. But adaptation in some areas is increasing.
There was an interesting article about the American film industry in the New Yorker some time ago with the statement: “Everyone is using it, nobody is talking about it.” Nevertheless, big names are positioning themselves: James Cameron is on the Board of Directors of Stability AI. Disney has entered into a partnership with OpenAI, initially only for user-generated content. The rate of adaptation is steadily increasing.
The first series to use AI on a large scale for Netflix is The Ethernautan Argentinian production. Entire VFX sequences were generated using AI for the first time.
According to Ted Sarandos (CEO of Netflix), these scenes cost a tenth of the usual production costs. Netflix has imposed a strict set of rules on its producers,
but these are the first productions with AI on this scale.
In the animation sector, it has been announced that OpenAI, in cooperation with the studios Vertigo Films and Native Foreign, would like to present an animated film in Cannes in 2026. For a fraction of the time and cost of a “conventional” production. We’ll see what comes of it. That’s a lot of marketing again. But you can see: AI is finding its way more and more into the production process.
AI has been playing a role in post-production for some time now. Especially in the field of visual effects, there are many starting points that are already being used. It hasn’t been talked about so much for a long time, but that is now changing. AI is also very important in pre-production.
There are two sides to the coin here: the big studios are using AI primarily to cut costs and achieve efficiency gains. This is the more threatening development for existing jobs. These studios have the money to invest in research and development. On the other hand, there are AI-native studios that not only optimize pre- and post-production, but also generate moving images as the primary form of production. The industry will settle somewhere in between.
But with the developments of the last few weeks, we may be witnessing the next big leap. What ByteDance, TikTok’s parent company, is showing with Seedance is moving in the direction of a ‘full production engine’. And it is landing directly in the hands of millions of creators who are quickly learning how to use it.
If a tool actually starts to suggest entire scene sequences, including editing, sound and lip sync, then that goes straight to the heart of classic post-production. We are then no longer just talking about ‘help’, but about the fact that additional work such as rough cuts or alternative camera angles could be automated and thus restricted as a well-paid service. We have to keep a very close eye on this.
Is there an example of something that many people have already seen where you can say that this is now possible with AI?
In The Irishman the de-aging processes were very expensively produced. Five years later, in Here with Tom Hanks, precisely this ageing process was AI-generated. It was played out on set in real time. You could see the younger faces of the actors in the camera.
Fully AI-generated content in the classic sense is still rarely seen at the moment. There are many short films and so-called spec spots, i.e. demo films made with small teams to show what is possible with AI. Animated films will also follow in the near future. But all of this raises big questions. That is the next big variable: How will copyright disputes develop? The last word is far from being spoken, but we will see a lot in the coming months.
What happens in the pre-production process and why can AI be used there?
Many large companies are already using AI in script analysis, data analysis and storyboarding. The more precisely you can plan a film in advance, including visually, the smoother the processes on set and the more efficient the result.
Although AI is still a long way from being able to write scripts, it is already possible to have individual text passages generated by AI. AI tools will soon be standard, especially in planning, location scouting, financing and costing. However, it is not a topic that the industry always talks about openly.
Netflix recently published rules on the use of AI.
The main point is that there is an obligation to notify AI. Producers are responsible for ensuring that the material supplied complies with legal requirements. It is also stipulated that AI cannot assume authorship. This shows that The responsibility still lies with the producers. Anyone using AI must ensure that everything is legally clean. That doesn’t make it any easier.
Which companies in the film industry suffer from the use of AI and which can benefit?
Small companies are quicker to adapt. If I were to set up a new company now, I would make my processes AI-compatible right from the start. This is a great opportunity: data analysis, predictive analytics, AI-based script evaluation. At the moment, smaller companies often lack access to relevant data. However, this could be democratized, for example through synthetic or aggregated data. Those who can use it to better target films to their audience will find it easier to get financed.
The job market is like many sectors of the economy: Jobs are not necessarily being cut, but roles are changing. This is a challenge for large organizations. Certain job profiles can disappear. If the core of film production changes, some roles will no longer exist. This requires major change processes. Small companies are less likely to have these issues.
Small is just a bit more agile?
Exactly. They are simply less concerned with their own transformation. On the other hand, the big companies own the IP, the intellectual property. Disney has announced that it will be using AI around its films, primarily for marketing purposes. We will have to see how this develops. But if you have a powerful IP, you have influence. This remains a huge advantage for the big companies.
At the same time, copyright issues are an open point. The big companies are using AI, but are actively suing the LLM model providers because their training with copyrighted material is a threat to their business basis. These landmark rulings point the way forward.
“The more doomsday-scenario-like or negative talk there is about the topic, the more likely it is that AI will be used in a non-transparent way”
In other words, it hasn’t really been decided yet?
No, not by a long shot. It’s very important to talk openly about this in our industry, to share concerns and to avoid creating fronts. The more doomsday-scenario-like or negative talk there is about the topic, the more likely it is that AI will be used in a non-transparent way. There needs to be a proper discourse in which everyone is heard. The Production Alliance took a great step in May 2024 to negotiate a set of rules with Verdi. These are all good signs that the issue is being taken seriously.
However, it is important that the film industry does not isolate itself in the discussions. The AI race is being fought on a completely different stage, at a political level: mainly between the USA and China. There are much bigger issues at stake in the geopolitical competition. With such dimensions, individual sectors can quickly fall behind, so you have to be realistic.
A balancing act between not losing touch and not wanting to go along with everything.
Some discussions, including legal ones, could be decided in favor of the AI leadership of individual countries. Some administrations, for example, do not currently consider copyright to be the greatest good. Striking a balance here – not cutting ourselves off from innovation, but not betraying our own values and foundations – is not easy. But we should look to the future with a certain positivity. The whole thing cannot be turned back.
In this context, are there any noteworthy approaches from companies to develop their own models or is it more a case of using existing models?
In the USA, for example, Moonvalley has trained its own model, supported by VC capital.
We discussed whether there should be European models at a panel during the Munich Film Festival. The problem: as development progresses, so do development costs and server capacities. This is a battle that the entertainment industry is unlikely to win.
That’s why partnerships like Lionsgate and Runway are interesting, even if they show that even large film catalogs are often not enough. It is more likely that there will have to be pre-trained models, i.e. basic models that are then fine-tuned through extended training.
Of course, it is a big problem that the flagship models come from China and America. They are fed with their bias and stereotypes. Even the people who train these models are biased to a certain extent and reflect a certain world view. That’s why it’s very important to me in my work to bring in experts who can take a critical look and address the issue of bias and stereotypes.
In Europe, we have a cultural film heritage and our own language and visuality that these models may not reflect. Filter techniques or additional training must be used to create opportunities so that we don’t lose our visual and cinematic identity and become too dependent.
It is a difficult issue to say that we want a European model. That would only be conceivable with very strong political support and in a European association. I think we should rather fall back on basic models and then design them appropriately.
“If data is missing, cultural representation is missing”
The question would also be: Is there even enough data? How many Finnish films are there, for example?
That is the sore point. If data is missing, the cultural representation in the model is missing. We then don’t see Finnish or German nuances, but the average from Hollywood. But big Hollywood productions have long since become a viewing habit here in Germany anyway, which is not always good for our own film market.
But this is also an opportunity for the German and European film market: we can use these possibilities to realize Hollywood opulence in the future with the current or even lower budgets.
The question arises: What remains when opulence is democratized? When spectacular images become generic? Then the story counts again. And we should take great care to preserve our own identity and tell our own stories.
How threatening are the current developments?
It is very important to me that we see innovation and disruption as an opportunity. Of course, many people are worried that their role will no longer be sought after in the future. But I can see that new roles are emerging. In the USA, a studio recently advertised 17 new AI positions. Of course, there is also other competition on the market. What I’m seeing in the AI film community and the AI community in general is that people from all kinds of backgrounds are making a name for themselves.
The human element remains the core, that is not a cliché. Nuances are crucial. I think it’s going to be a hybrid production – also because the “Uncanny Valley”, that last, difficult ten percent to human-looking perfection, will be with us for some time to come. Even if the gap may have shrunk considerably with current developments, especially in terms of emotional depth and facial expressions.
But viewing habits will change, if only because social media is flooded with AI content. To be honest, when I watch a video of Sora 2, I can hardly distinguish it from reality without a watermark. That worries me with regard to our media reality.
But for the creatives, it is an impetus for further development. The Filmakademie Baden-Württemberg is doing this in an exemplary way: they convey a desire to create so that people embrace the technology and move forward. This development is coming with great force. Opposing it is the least effective way. You don’t have to approve of everything, but you have to understand it.
Understand, stay tuned, become curious…
The most important thing is to remain curious and really get to grips with developments. No one can predict exactly what will happen. But you can recognize patterns in the direction in which technology and opportunities are developing.
There are many examples that show this: AI can help us get back to the core of our work – telling stories – and free us from one or two administrative superstructures.
Interview: Sabrina Pohlmann
With our interviews, we want to introduce you to different perspectives and players in the field of AI. The positions of our interview partners do not necessarily reflect the positions of the ARIC.
Further interviews:
