Skip to content
Where does this figure come from?Uses20 October 2026

"A generated image is a phone charge"

Guillaume PakulaClément ReynaudSébastien PierfedericiAlexandre Veilly By Projet Celsius
Published on 20 October 2026 · 4 min read
An Nvidia Tesla compute card lying flat on a white background, with its black and green shroud

The comparison did the rounds of the media in late 2023. It came from a serious study, but from its preliminary version, from the most power-hungry generator, and from a phone charge counted as 12 Wh.

Contents
1.

Where the figure comes from

In November 2023, Sasha Luccioni, Yacine Jernite and Emma Strubell posted "Power Hungry Processing" online, the first systematic measurement of the electricity used by AI tasks: 88 open models, 10 tasks, 1,000 trials per task. In this first version, the most power-hungry image generator uses roughly as much as a smartphone charge, which the study then counts at 12 Wh. The press picks up the phrase and generalises it to all images.

2.

What the published version says

The version published in 2024 counts the charge at 22 Wh, the value used by the US Environmental Protection Agency at the time. The most power-hungry generator then amounts to half a charge, and the average across generators to 13%. The measurements themselves have not changed: 2.9 Wh on average per image, 1.35 Wh for the median generator, 11.5 Wh for the most power-hungry.

AI Impact, by Projet CelsiusUses

Electricity by type of AI task

For 1,000 inferences, in kilowatt-hours, logarithmic scale · Measurements published in 2024

Image generation2.9
Text summarisation0.049
Text generation0.047
Object detection0.038
Image classification0.007
Text classification0.002
0.0010.010.1110

Source: Luccioni et al., Power Hungry Processing, 2024 · projetcelsius.com/en/ai · CC BY 4.0

Image generation is the most expensive task in the study, 60 times as much as text generation.

3.

The comparison in 2026

The US agency now counts a charge at 19 Wh. An average image represents 15% of that, and an image from the most power-hungry generator 60%. Compared with a written question (0.3 Wh), an image costs about 10 times more. With the world average electricity mix, that comes to 1.3 gram of CO2e, the equivalent of 9 metres in a petrol or diesel car.

Other measurements confirm the order of magnitude: MIT Technology Review measured 0.63 Wh for an image of 1,024 pixels a side with Stable Diffusion 3 Medium, and 1.22 Wh with twice the number of computation steps.

Electricity for an AI-generated image: the formula and the measurement
The formula, late 20231phone charge
  • the most power-hungry generator
  • the preliminary version of the study
  • a charge counted at 12 Wh
False
The measurement, in 20262.9Wh per image, on average
  • 15% of a 19 Wh charge
  • 10 times a written question
  • 1.3 gram of CO2e with the world average electricity mix
Electricity used by an AI-generated image→
4.

What the figure does not say

No commercial generator, whether OpenAI's, Google's or Midjourney, publishes its consumption per image. All the available values come from open models measured by researchers. Image resolution and the number of computation steps make the energy vary by a factor of 1.3 to 4.7: the value for a given image can differ greatly from the average.

Point to note

A phone charge is not a fixed standard: its official value went from 12 to 22 and then to 19 Wh in 2 years. For comparisons, it is better to give watt-hours.

Guillaume PakulaClément ReynaudSébastien PierfedericiAlexandre Veilly
By Projet CelsiusAI Impact, Projet Celsius · 20 October 2026
Write to us →

Sources

3 documents

Every figure was read in its source; the hub's methodology is described in Methodology and sources.

Cite this analysis

AI Impact, Projet Celsius, "A generated image is a phone charge", 20 October 2026. https://projetcelsius.com/en/ai/analyses/image-equals-phone-charge/

Charts and maps

A question about a figure?

An error, a source to add, a methodology question, a press request: write to us. Guillaume Pakula, Clément Reynaud, Sébastien Pierfederici and Alexandre Veilly will reply.

Write to us →

Projet Celsius, which runs this hub, also helps companies measure and reduce the footprint of their AI use.

Accounting for AI in your Bilan Carbone® ↗
Projet Celsius, environmental strategy consultancy

We support companies that account for AI in their Bilan Carbone®

Visiter Projet Celsius ↗