Computational footprint is the resource requirement of operating text-to-image models, including large datasets, powerful hardware, substantial energy use, and environmental cost. section 1.6 section 13.2
The footprint appears across the model lifecycle:
- Training and deployment demand significant computational resources and energy. section 13.2
- Greater resource demand creates scalability and environmental challenges. section 1.6
- Pruning, quantization, and transfer learning can reduce costs, but preserving output quality remains difficult. section 1.6 section 13.2
Model scalability
A state-of-the-art text-to-image model may require large datasets and powerful hardware for training and deployment, making lower-cost operation without quality loss an ongoing research problem. section 13.2
