New report explores path to better understand and manage AI video emissions

Media companies and production teams lack the consistent data and methodologies needed to measure the carbon impact of AI video generation in production, according to new research unveiled today by the Carbon Trust.

'The carbon impact of AI video generation' was commissioned by DIMPACT - a coalition of leading media organisations including the BBC, Netflix and Spotify. Developed with input from DIMPACT participants, the report sets out how AI video emissions can be better understood, measured and managed.

The report identifies AI video generation as one of the more energy intensive generative AI applications, requiring at least two orders of magnitude more energy than responding to a simple text query according to recent evidence.

For one specific AI model, a 5.4-second AI-generated video at 720p can have a lifecycle carbon impact of around 50-100 gCO₂e per video on the US electrical grid - including embodied emissions, training, and inference. This result is highly sensitive to the particular model used, the associated training emissions and video settings, yet is broadly comparable to the lifecycle emissions of making one cup of coffee from a coffee pod. To further contextualise this figure, the report's case study found that a real-world application of AI video generation required over 2,000 videos to produce a short scene for a streaming series.

The case study demonstrates some indications of use cases which could support energy efficiency improvements over traditional production methods, but existing data and measurement approaches introduce uncertainty to the conclusions and further exploration is needed to understand the carbon impact for a broader range of use cases.

One core objective of this research is to facilitate increased data transparency by establishing a consistent measurement framework for AI usage. This is important because one analysis included in this study notes that 84% of AI usage today comes from models with no environmental disclosure, while only 2% came from models with direct disclosure.

Despite the lack of transparency, the report still points to practical opportunities for progress and provides recommendations on how digital media companies can begin to monitor and manage carbon impacts despite uncertainty.

This research is a combination of detailed literature reviews from existing studies, culminating in a measurement framework for calculating AI emissions and an estimated carbon footprint figure for general video generation, as well as an industry-specific case study highlighting the relative impacts of AI tools in film production vs traditional methods.

The report recommends active collaboration between model developers, data centre operators, model providers, LCA practitioners, the scientific community and users to develop consistent lifecycle assessment methods in the form of product category rules (PCR) for generative AI uses, including video generation. With consistent rules in place:

• Model providers can confidently disclose comparable emissions associated with using their services
• All parties can identify and implement opportunities to reduce these emissions
• Users can make informed decisions about the relative carbon impact of using generative AI in video generation workflows

Bob Burgoyne, Data Centre Lead at the Carbon Trust said:

"Generative AI is moving into media production and content creation faster than the industry can fully measure its climate impact. This research is a step towards building the data, transparency and shared understanding needed to enable organisations to make more informed decisions as the technology continues to evolve."

Glynn Roberts, Global Director at SLR Consulting and Executive Sponsor of DIMPACT said:

"We commissioned this research to bring clarity to a fast-moving space where decisions are increasingly being made without consistent data. As AI becomes embedded in media production, it is critical that the industry aligns on how we measure and manage its environmental impact."

Notes to editors

About the Carbon Trust

The Carbon Trust is a global climate consultancy driven by the mission to accelerate the move to a decarbonised future. Its vision for the ICT sector is for digital infrastructure and energy systems to evolve together for low carbon growth and to accelerate the energy transition.

From power procurement and grid interaction to system wide measurement and circular hardware strategies, its experts know what low carbon data centre growth looks like in practice. Working directly with owners, operators and major customers, the Carbon Trust supports data centre projects to scale in ways that reduce overall system emissions, rather than shifting them elsewhere in the supply chain.

About DIMPACT

Facilitated by SLR Consulting, DIMPACT is a 'think and do' coalition working to align industry changemakers and policymakers around meaningful, science-based solutions that reduce the environmental impacts of serving digital media products. DIMPACT convenes and unites industry leaders and changemakers, sharing research, resources, and best practices to catalyse collaboration and accelerate action.