Generative AI models for image, audio, and video are advancing, delivering more realistic and creative content that is becoming more controllable over longer sessions. Although studios may have been quick to experiment with gen AI content creation, they will likely be more cautious in moving it into full production. Some reasons for this include the immaturity of the tools and the challenges of content creation with current public models that may expose them to liability and threaten the defensibility of their intellectual property (IP). However, there is growing belief that gen AI applied across their businesses could help studios reduce costs and grow their profitability.
Indeed, leading studios are facing cost pressures, and very few are showing profits.1 Revenues are high, but operating expenses and the costs of production, marketing, and advertising have typically become higher.2 This is often true for many studio streamers that are funding their streaming services without profit while losing revenues from declining cable TV subscriptions and advertising. Inflation, higher interest rates, and the impacts of the COVID-19 pandemic have further inflated costs, and studios now also compete with social media, user-generated content, and video games for consumer attention and revenues.
In 2025, Deloitte predicts that the biggest TV and film studios—especially those in the United States and European Union—will be cautious in adopting generative AI for content creation, with less than 3% of their production budgets going to these tools.3 But we also predict that operational spending will shift about 7% into emerging generative-AI-enabled tools supporting functions like contract and talent management, permitting and planning, marketing and advertising, and localization and dubbing of content that can expand their reach into diverse global markets.
This approach can help studios slow the potential disruptions that gen AI can pose to talent and content, while more quickly adopting gen AI tools that can help reduce costs and accelerate performance across their businesses. However, independent content creators and social media platforms are moving quickly to adopt gen AI into their workflows and content, potentially enabling new forms of media to emerge that could further disadvantage traditional studios competing for scarce attention time.4
The availability of cheap, off-the-shelf large language models (LLMs) and diffusion models have helped enable studios to experiment with rapid prototyping of scripts, dialog, and story elements, and with early visualization and discovery of character and set design.5 Some studios are using generative tools to de-age their celebrities or create digital twins that can be lent to commercials—or to post-mortem productions.6 In such cases, studios can help control for potential liabilities by writing protections directly into the contracts with actors. The coming year will likely see more third-party production groups selling services and tools to studios offering such capabilities.
While content creation with generative AI can enable greater creativity in preproduction, it cannot yet deliver Hollywood-level productions.7 Although the strongest visual diffusion models are now able to generate photorealistic images, their outputs still seem “uncanny”—too hyper-realistic.8 Leading video models can generate short clips, but cannot produce longer, more coherent stories.9 Although video-generation models are advancing quickly, it may still take time before they are mature enough to integrate into existing tools and production pipelines.
The year ahead will likely see independent creators leading the way in content creation with generative AI.
However, these limitations may be fine for social media creators, who are often incentivized to create and publish quickly. Fast-paced quick cuts have gained popularity, though this could be changing.10 Social video lengths are often short, and liabilities are perhaps less concerning. Some early adopters of generative models and tools are regularly publishing their experiments on social media, showcasing the fast-moving advances of video models teased out of third-party solutions.11
The year ahead will likely see independent creators leading the way in content creation with generative AI. This could help studios defer their own risks while they watch to see how the capabilities evolve. But it could also cede more attention time to user-generated content platforms that are becoming highly competitive with traditional media.
Large studios are also concerned about how gen AI content tools may raise their exposure to IP and liability risks, or make their own generative content indefensible as original works.12 Some of the most capable publicly available models have been trained on public data, like images and videos from other creatives, making their outputs fundamentally derivative.13 If a studio uses outputs from a public model for profit, and that model includes the protected works of other artists in its training set, the studio could be held liable for infringement. With potentially billions of works in a training set, infringement may be nigh impossible to prove— finding a “drop of water in the ocean”—but this uncertainty may be enough to scare off studios whose livelihood is making, securing, and defending their IP. Independent artists and creators are already suing public models for perceived infringements of their works in training sets,14 as are publishers15 and music labels.16
Public models could also make it difficult or impossible for studios to secure their own IP. The US Copyright Act has required sufficient “human authorship” before it will issue a copyright.17 In recent considerations, the US Copyright Office acknowledges that the degree of human authorship in works that include AI or generative inputs can vary on a case-by-case basis, and such works can receive a copyright provided they meet requirements of “sufficiency.” Which is to say, studios can get copyrights for human works that are supported by generative AI tools, but only to a degree and not for works primarily produced by generative models. This is an ongoing discussion in a maturing space, but the lack of precise definitions introduces further uncertainty and risk.
Hungry for more data to feed their training sets, leading gen AI providers have been courting studios and incentivizing them to license their content archives.18 However, studios may resist this entirely since their IP is their livelihood, or they may charge onerously high rates to gen AI companies that may already be straining under their own operational costs. Studios could even see an advantage in collectively denying data to training sets in hopes that they might inhibit frontier models—the algorithms being encoded and trained to generate text, audio, image, and video.
Additionally, studios—especially those in the United States—must work with guilds and labor unions that have shown strong resistance to adopting generative AI and have extracted guarantees from studios limiting its use.19 Similar labor pushback is emerging in the United Kingdom20 and in the European Union where studios will also need to be compliant with regulations like the EU Artificial Intelligence Act governing the safety of models, and the General Data Protection Regulation governing how they collect and store data that may be used for training.21
In Deloitte’s 2024 TMT Predictions, we discussed the rise of private gen AI models to help avoid some of the challenges with public models and to gain more control over outputs.22 Studios could avoid the liabilities and copyright challenges of public models by training their own models on their own IP.
But training generative models has become very expensive—around US$100 billion to train a leading-edge model, by some estimates23—and the costs of inference and retraining can grow with usage. Open-source solutions (often more accurately referred to as “open weight”) may defray some costs, but their training sets are opaque, and costs are still high.24 Studios may be challenged to attract expensive technical talent able to build such models—talent that may be more inclined to work with hyperscalers able to pay premium salaries. Additionally, investing in today’s models could require updates within six months due to the rapid pace of model development. To build more effective private models, studios and investors may have to think and act more like tech companies, building and maintaining ecosystem relationships with—and paying rents to—tech providers. For these reasons, studios may be less likely to train their own models without considerable shifts in economics.
However, the year ahead could see a flurry of partnerships between studios and providers that could share the cost burdens more equitably.25 In such partnerships, a third party could provide a pretrained model and interface that can then be further trained and customized with content owned by a studio. The model could then deliver generative content that follows the aesthetic of a studio, for example, or includes their signature characters and set pieces. Additionally, studios might be able to control against potential IP concerns by showing derivations from their own content. Still, providers of such capabilities could be expected to show greater maturity of the tools and outputs.
In the coming year, studios are expected to experiment with generative AI content creation, but they will likely move more quickly to understand how gen AI can better enable and optimize more of their business. Generative AI may be able to help automate and augment contract negotiations, talent and workforce management, finance and accounting, and media operations like localization, marketing and promotions, and storage and distribution.
Studios are likely to absorb some of these capabilities through the software and software-as-a-service solutions they already use. Smaller companies are also emerging to tackle time-consuming—and costly—parts of the preproduction process. Generative AI can accelerate script evaluation, break them down and assign them to production schedules, and even “scout” for potential locations that could support the script.26 Generative AI could also unlock archives of content, for example, by “watching” old films and tagging them for actors, themes, and moods. This could then help enable streamers to dynamically resurface and monetize old content to meet more personalized recommendations or trending moments.27
To help accelerate and amplify content distribution, some studios are now leveraging language and voice models that enhance translation and dubbing so it can reach broader global audiences.28 These tools enable high-fidelity voicings that can be highly expressive and emotional, fine-tuned by users.29 This could be a boon for content creators and distributors that are addressing global markets, both in exporting content to them and importing it from them. Leading platforms for user-generated content creation have also extended these capabilities to their creators.30
Gen AI dubbing and translation could also help enable greater sharing of cultures, potentially generating large hits that might otherwise remain as local phenomena. Analysis of the responses to Deloitte’s 2024 Digital Media Trends survey showed that 66% of Americans surveyed enjoy watching TV shows or movies that help them learn about cultures different from their own.31 Generative AI could not only help media companies grow their margins and compete more effectively, but it could also bring audiences closer together.
Smaller companies are also emerging to tackle time-consuming—and costly—parts of the preproduction process.
Like most companies, studios, streamers, and creative talent are both fascinated and concerned about the capabilities of gen AI. For studios exploring generative AI content creation in the year ahead, the dominant driver of adoption will likely be the magic and creativity it seems to enable—the strange fever dream of frontier models remixing human creativity into new forms. They will also be driven by the concern that new forms of media could emerge from outside the Hollywood ecosystem.
More independent content creators are demonstrating what can be done with the latest synthetic media capabilities rapidly entering the market. Hollywood studios once enjoyed controlling the scarcity of content and distribution, but now these are both abundant and democratized.32 In the year ahead, the sense of looming content disruption will likely only grow.
Every month or so there are new developments in frontier models advancing their capabilities further along the path towards human intelligence, creativity, and insight. A year ago, it was thought that, by 2030, a major blockbuster film would likely be generated almost entirely from AI.33 In 2025, that lofty goal may seem a bit more achievable.
In the meantime, content owners will likely work to shore up the competitive moat around their IP, pursuing more litigation against, and regulation of, public models for perceived copyright violations. Regulators could require leading model providers to prove their training sets do not infringe on pre-existing content rights. Most big studios will likely resist offers from model providers to license their content catalogs into their public training sets, likely preferring to partner with smaller companies able to build more bespoke and protected models around studio IP—if the economics are favorable.
At the macro level, generative AI has required enormous capital intensity that could slow growth if a path to broad economic value isn’t revealed within the next year or so.34 (See this year’s TMT Prediction about on-device generative AI.) However, if the next generation of frontier models can overcome existing challenges, capabilities could advance quickly. Efforts will likely emerge to reduce the costs to train and run models, and to help reduce the amount of data needed.
Big studios are also large enterprises that will likely adopt more gen AI capabilities that are focused on cutting costs, optimizing their businesses, raising productivity, and expanding and accelerating their reach to customers. In Deloitte’s 2024 State of Generative AI in the Enterprise survey, 42% of executives surveyed report efficiency, productivity, and cost reductions as their single most important benefit achieved from using AI; and 58% reported a range of other benefits such as increased innovation, improved products and services, or enhanced customer relationships.35 There appears to be growing interest in applying these capabilities to modern businesses.
A rising tide lifts all boats, as the saying goes. Gen AI tools seem poised to help more smaller companies and creators to achieve the kinds of productivity and levels of quality once reserved solely for the largest companies. Smaller studios and independent creators could become much more capable, while still being relatively free of the risks and cost overheads born by larger studios. The biggest studios among them may need to lower their costs and accelerate their time to market if they hope to compete—not just with each other, but also with user-generated content platforms, social media, and gaming. Production and distribution may be less scarce, but attention remains a finite resource.