What is that worth
in AI images?
Environmental figures are hard to picture. Compare familiar activities with the estimated energy use or carbon footprint of generating AI images.
Estimated energy
400 Wh
AI-generated images
Typical estimate uses 2.9 Wh per image. Because a more energy-intensive image has a larger impact, fewer images fit into the same equivalent.
It started with a slightly cheeky question.
Discussions about AI image generation often include figures for electricity use or carbon emissions, but isolated numbers are difficult to interpret. Is 3 Wh significant? What does 500 g CO₂e feel like in everyday terms?
AI Image Equivalent puts those estimates beside familiar activities: running a fan, boiling a kettle, driving a car or taking a flight.
The purpose isn’t to excuse AI’s environmental impact or exaggerate it. It’s to replace abstract numbers with understandable context, using transparent assumptions and published evidence.
ONE NUMBER ISN’T THE WHOLE STORY
Same image.
Different impact.
Model choice, hardware, resolution, inference steps, batching, data-centre efficiency and the local electricity mix can all change the result. That’s why every answer includes a range.
Transparent by design.
These are estimates for relative scale, not measurements of any individual image, appliance or journey.
01 AI image generation+
There is no universal energy cost. The central 2.9 Wh estimate follows the mean measured by Luccioni, Jernite & Strubell across image-generation models. We use 0.7 to 11.5 Wh to expose variation from efficient small models through the study’s high observed result. Inference only; hardware, resolution, steps, batching and data-centre overhead can change real use.
02 Electricity & carbon+
Watts describe power. Watt-hours describe energy over time. One kWh equals 1,000 Wh. Carbon is calculated separately: energy × 143.95 gCO₂e/kWh, the 2026 UK grid generation plus transmission/distribution factor. The same image on another grid can have very different emissions.
03 Flights+
Simple mode estimates distance at 800 km per flight-hour. Because the calculator does not know the route, it automatically uses 3,700 km as a transparent distance-only proxy for short or long haul. Official UK factors classify haul by destination region, so this is an approximation. Government passenger-km factors include an 8% distance uplift. Cabin class changes passenger allocation. The optional radiative-forcing variant represents additional non-CO₂ effects; UK guidance uses a 1.9 multiplier while noting substantial scientific uncertainty.
04 Vehicles & transit+
Cars use 2026 average-size UK factors and are divided by passengers entered. Public transport is already expressed per passenger-km. EV carbon follows the UK grid; the energy view uses 0.205 kWh/km. Actual vehicle, load, route, driving and occupancy matter.
05 Home heating+
Electric heat is compared by input electricity. Gas uses 0.18231 kgCO₂e/kWh (gross calorific value). Heat-pump output is contextualised with an illustrative coefficient of performance of 3.2; weather and system design alter real performance.
Measured 1,000 sequential inferences across popular image-generation models. The study reports a 2.907 kWh mean, a 1.35 kWh median and an 11.49 kWh upper observed model result per 1,000 images.
Open source ↗2026UK Government GHG Conversion Factors 2026July 2026 revised flat file. Used for UK electricity, fuels, vehicles, public transport and aviation.
Open source ↗2024DfT journey emissions methodologyExplains distance uplift, passenger allocation, cabin class and the optional 1.9 radiative-forcing multiplier.
Open source ↗Questions, corrections or source suggestions?
Found an outdated assumption, spotted a calculation issue or have a question about the methodology? We’d like to hear from you.
ATKNProjects@proton.me ↗