Sustainable AI Group develops the first empirically-validated method to estimate the energy footprint of AI video generation

Generative AI is increasingly used across industries, and its modalities continue to expand. While the energy cost of text generation is on its way to becoming better understood, the impact of video generation has remained largely unknown. That gap matters more by the day: AI videos are being built into major consumer platforms, and the advertising and communication sector is adopting it at scale.

Part of why that gap persists is that the providers of the largest closed models have so far declined to share the data needed to measure their footprint directly. That transparency must be demanded, and SAIG continues to call for it. But until it arrives, the responsible path is to use the best available methods to quantify these impacts — because what cannot be measured cannot be reduced. Credible estimates are what allow more sustainable models to be distinguished from less sustainable ones, and what ultimately raise the bar for the industry as a whole.

SAIG's methodology is built for exactly that situation. Rather than relying on access to a model's weights or size, it estimates energy consumption from observable generation parameters — resolution, duration, and architectural principles — which makes it possible to assess open and closed models alike, even when their developers disclose nothing. A scientific paper detailing the methodology and full results has been submitted for peer review and will be released publicly soon, alongside a companion piece discussing the findings.

Among the headline findings the paper will detail:

• A few seconds of AI-generated video can consume more than 1,000 times — and in some cases close to 2,000 times — the energy of a single text prompt.

• Across the models and settings tested, per-video energy spans nearly three orders of magnitude; even at an identical task, the most and least efficient models differ by roughly an order of magnitude.

• Validated across multiple open-source models and hardware configurations, the framework predicts energy consumption to within a few percentage points.

The research was conducted by SAIG, which developed the energy-estimation methodology for video generation now submitted for scientific peer review. The methodology was then operationalized and made usable in an enterprise setting through integration with Code Carbon's EcoLogits and e-footprint created by Publicis Sapient France and open sourced within BoaVizta.

This work is the next step in a growing body of methodologies that SAIG is building to measure AI's footprint across modalities — with more to follow.

"Video is the fastest-growing and least understood modality of generative AI, increasingly being used across social media and advertising alike, yet until now its footprint has been almost impossible to measure. We thank Publicis and the GenAI Footprint Alliance for helping us build the knowledge to understand, compare, and ultimately reduce the environmental impact of AI." — Sasha Luccioni, Co-founder and Chief Scientific Officer, Sustainable AI Group

"Limiting our industry's impact on the environment is a priority. Since the acceleration of AI, we have been facing gaps in our ability to measure it. Our CSR teams and Publicis Sapient in France conceived this initiative to deliver a complete estimate of the environmental footprint of our content, including AI. We are proud that our alliance now brings together some of the largest international companies." — Agathe Bousquet, President, Publicis Groupe in France

"At ENGIE, we are very proud to have been one of the first Groups to join this pioneering initiative. As a leader in the energy transition, ENGIE provides its clients—particularly data center operators—with sustainable energy solutions based on renewable electricity generation. It is also our responsibility to measure the greenhouse gas emissions associated with our own use of AI, both in our service and industrial activities. ‘GenAI footprint’ will enable us to benefit from the best available expertise to establish the most accurate assessment possible and anticipate action plans.” — Florence Colombo-Fouquet, Group ESG Vice President, ENGIE

About Sustainable AI Group

Sustainable AI Group (SAIG) is a research and advisory firm helping enterprises measure, compare, and act on the environmental impacts of AI. The firm was founded by two pioneers in the field: Dr. Sasha Luccioni, a renowned AI researcher with a decade of experience building field-defining benchmarks, and Boris Gamazaychikov, a strategist specializing in sustainable AI enterprise implementation. Together, they help organizations make AI sustainable in practice by grounding deployment decisions in empirical evidence. SAIG supports organizations adopting AI by conducting rigorous studies on fundamental environmental questions, developing evidence-based procurement guidelines, and engineering practical tools that bridge the gap between climate science and real-world corporate decision-making. Learn more at: sustainableaigroup.com