Researchers have discovered that the carbon footprint of generated AI-based tools that can turn text prompts into images and videos is far worse than they previously thought.
As detailed in a new paper, researchers on open source AI platforms have found that when face-hugging doubles the length of the generated video, the energy demand for intertext generators is four times higher.
For example, a 6-second AI video clip consumes four times the energy of a 3-second clip.
“These findings highlight both the structural inefficiencies of current video diffusion pipelines and the urgent need for efficiency-oriented design,” the researchers concluded in their paper.
Experts warn that they are deploying generative AI tools without a full understanding of their true environmental impact.
“In the end, we found that the general understanding of AI's energy consumption is full of holes.” MIT Technology Review I wrote about it in a recent analysis.
The image generator produced a single 1,024 x 1,024 pixel image using the equivalent of 5 seconds of microwave warming, but the video generator proved to be much more energy-intensive. To spit out a 5-second clip, researchers found that it was equivalent to running the microwave for more than an hour. As the length increases, if they consume far more power, math doesn't look good.
These demands will rise even faster with longer clips, meaning “hardware and environmental costs will increase rapidly and environmental costs will increase.”
Fortunately, there are ways to slim down those requests, such as intelligent caching, reuse of existing AI generations, and “pruning” sift through inefficient examples of training data sets.
However, it remains to be seen whether these efforts will be sufficient to plunge into the enormous power consumption of current AI tools. Recent research shows that AI-related energy usage already accounts for 20% of global data center electricity demand, resulting in a larger scale of its impact and already represents AI-related energy usage.
Meanwhile, the tech giant has invested tens of millions of dollars in building infrastructure, and sometimes abandoning its climate targets in the process. In its 2024 Environmental Impact Report, Google acknowledged that it was behind its ambitious plan to reach net zero carbon emissions by 2030.
Earlier this year, the company released the VEO 3 AI video generator, which later boasted that users had created over 40 million videos in just seven weeks.
The environmental impact of the tool remains unknown, but Google is not exactly encouraged to investigate its substantial contribution to carbon emissions, but it could be far worse than we think.
Details of AI energy usage: How much electricity does it actually take to use AI?
