Energy consumption is the electrical energy required by computing processes. In text-to-image generation, training and deployment consume substantial energy because these systems depend on large datasets and powerful hardware, creating environmental costs. section 1.6 section 13.2
Energy consumption becomes a practical design concern through three linked pressures:
- Training state-of-the-art models requires significant computational resources, including large datasets and powerful hardware. section 1.6 section 13.2
- Pruning, quantization, and transfer learning can reduce the computational footprint, but they do not remove the scalability challenge. section 1.6 section 13.2
- Reducing energy use while preserving output quality remains an active research trade-off for text-to-image systems. section 1.6
A projected scaling risk
A June 2024 Computer Weekly report warned that, without intervention, continued AI growth could account for half of the world’s total energy supply by 2040. section 1.6
