Generating AI videos can cause more problems than deepfakes, new research warns of the looming global crisis

A calm reality is emerging as AI-generated videos flood social feeds and creative platforms. The biggest threat may be energy itself, not misinformation or deepfakes. A new study by embracing the titled face Video killed the energy budgetreveals that the text-to-video system consumes an incredible amount of power and costs a surge as the clip gets longer and sharper.

Energy becomes uncontrollable in spirals

Unlike text and image generation, video generation is exponentially expensive for the Earth. Researchers found that the energy and latency scale, along with video resolution and length, is secondary and linearly in the removal step. In reality, a 6-second clip may require four times more power than a 3-second clip. For high resolution videos, consumption shoots up to hundreds of watts per clip. This is thousands of times larger than generating text or images.

AI Power Hanger

This revelation reflects warnings from industry veterans. Former Google CEO Eric Schmidt recently argued that the natural limitation of AI is electricity, not silicon chips. It predicts that the US will need to be equivalent to 92 nuclear power plants to burn AI ambitions. Already, Microsoft has signed a nuclear energy trade, and Openai has invested in Fusion Startups to secure its strength for future growth.

Why is it more important than deepfake?

Society is obsessed with the dangers of deepfakes, but experts warn that the environmental costs of generated videos can cause their own crisis. The embracing face team observed that GPU usage accounts for more than 80% of the energy in all models tested, with larger systems consuming 3,000 times more power than lighter systems. On a large scale, this demand could derail global climate commitments, as seen in Google's 2024 report, primarily showing a 13% increase in carbon emissions driven primarily by AI.

The embracing face report suggests strategies such as diffusion cache, pruning inefficient training data, and quantization to reduce carbon footprint. However, these fixes may only slow the tide down. As Sasha Russioni, who hugged her face, said, “Video spreading is much more costly than text or image generation. It emphasizes the need for hardware-conscious optimization and sustainable model design.”

AI videos promise creativity without limitations, but have an invisible price tag. Unless efficient innovation keeps up, technologies that drive digital imagination can be too expensive to pay in the face of environmental bills. The upcoming crisis may be not only about what AI creates, but how much power it will take to create it.