Image generation models are more power-hungry than text generation models, study says

A team of researchers from Carnegie Mellon University and Hugging Face recently published a study that showed that image-generating models are more power-hungry than text-generating models.

The team is interested in different types of generativeAIThe energy consumption of the models, including chatbot, text generation model, and image generation model, was measured. It was found that the energy consumption of chatbots and text generation models per thousand interactions is relatively low, equivalent to 161 TP3T for a single charge of a cell phone.However, the energy consumption of the picture generation models per thousand interactions reaches a staggering 11.49 kWh, which implies that they consume more power than the other types of models.

The researchers also noted that the energy consumption of these image generation models depends on the size and quality of the images they generate. The wide variation between the performance of different image generation models suggests that more research is needed to understand and balance the environmental costs and theartificial intelligence (AI)Modeling Gains.

The findings of this study are important for organizations that use generative AI models. As more organizations begin to adopt generative AI technologies, they need to be aware that different types of models have different environmental impacts. If organizations want to reduce their environmental impact, they should consider using more efficient models or optimizing model settings and parameters to improve energy efficiency.

In addition, the results of this study emphasize the importance of sustainability in AI development. Future research should aim to develop more environmentally friendly AI models and techniques to minimize their impact on the environment and contribute to sustainable development.