Environmental impact describes how text-to-image systems affect the natural environment through resource use, hardware demands, and energy consumption during training and deployment. section 1.6 section 13.2
Text-to-image models create environmental pressure through several linked mechanisms:
- Training and deployment require large datasets, powerful hardware, and substantial energy. section 1.6 section 13.2
- Pruning, quantization, and transfer learning can reduce the computational footprint, but scalability remains difficult. section 1.6 section 13.2
- Maintaining output quality while reducing computation remains a central research challenge. section 1.6 section 13.2
Scale warning
A June 2024 projection warned that unrestricted growth in AI use could consume half of the world’s total energy supply by 2040, illustrating why model efficiency matters at scale. section 1.6
