Digital
Generating an AI image
Tekoälykuvan generointi
0.0
Truecost score
Data confidence:
HIGH
Carbon well-measured by multiple peer-reviewed studies. Water has moderate uncertainty.
Generating a single AI image consumes approximately 0.002–0.007 kWh of electricity, resulting in roughly 2–7 grams of CO₂.
Did you know? Generating 1,000 AI images produces less CO₂ than a single 20km car trip.
Transparent calculation
How was this number determined?
The Truecost score is calculated from absolute physical values. Each row below shows the measured value, how it was normalized, and where it comes from.
| Dimension | Absolute value | Score 100 = | Normalized | Weight | Weighted | Confidence |
|---|---|---|---|---|---|---|
| Carbon Emissions | 0.005 kg CO₂e | 9.5 kg CO₂e | 0.05 | ×0.35 | 0.02 | HIGH |
| Water Consumption | 1.0 liters | 3840 liters | 0.03 | ×0.2 | 0.01 | MEDIUM |
| Land Use | 0.0 m²·year | 51 m²·year | 0.0 | ×0.2 | 0.00 | HIGH |
| Waste | 0.0 kg | 5 kg | 0.0 | ×0.15 | 0.00 | HIGH |
| Toxicity | 0.0 µDALY | 162 µDALY | 0.0 | ×0.1 | 0.00 | HIGH |
| Truecost score (weighted sum) | 0.0 | |||||
Share of your annual planetary budget
Carbon Emissions
<0.01%
Water Consumption
<0.01%
Source data by dimension
Where do the absolute values come from?
Carbon Emissions
HIGH
300W GPU × 10s = 0.00083 kWh × 3 (PUE overhead) ≈ 0.0025 kWh. Safety margin 2x → 0.005 kWh × ~1 kg CO₂/kWh = 0.005 kg CO₂e
- Luccioni et al. (2023): Power Hungry Processing — energy and carbon costs of generative AI inference
- IEA (2024): Electricity consumption of data centres, AI and crypto
Well-measured. Range 0.002–0.010 kg CO₂e depending on model size and hardware.
Water Consumption
MEDIUM
Data center WUE ~1.8 L/kWh × 0.005 kWh ≈ 0.009 L direct. Rounded up to 1L to include cooling lifecycle.
- Li et al. (2023): Making AI Less Thirsty
Varies significantly by data center location. Arid regions much higher.
Land Use
HIGH
Data center land per inference: <0.001 m²·year
Waste
HIGH
No physical waste from a single inference
Toxicity
HIGH
No measurable toxicity from a single inference
Comparisons
- ~600x less CO₂ than a 20km car trip to a photo shoot
- Equivalent to charging your phone once
Methodology
Based on GPU power consumption (300W for ~10 seconds inference) and average grid carbon intensity. Water impact from data center cooling estimated via WUE metrics.
Sources
- Luccioni et al. (2023): Power Hungry Processing
- IEA (2024): Electricity consumption of data centres, AI and crypto
- Li et al. (2023): Making AI Less Thirsty