Skip to content
Get in touchFR
Updated in October 2026
Explained

What is the environmental impact of an AI-generated video?

This article looks at what an AI-generated video amounts to in CO2, electricity and water, based on measurements published in 2025. A simulator shows the effect of duration and resolution, and the article also covers the gap between generators and the origin of the "1 kWh per video" figure.

Guillaume Pakula
By Guillaume Pakula, co-founder of Celsius. Since 2020, he has helped 80+ organisations with their Bilan Carbone® and climate strategy.
October 2026
Updated October 2026 · 7 min
A 5-second AI-generated video emits about 41 gCO2e, equivalent to driving 290 metres and to the emissions of 31 AI-generated images. A video 2 times as long needs 4 times as much energy. Depending on the generator and its settings, measurements range from 0.06 to 190 gCO2e, and none of them covers a commercial service: the "1 kWh per Sora video" is an estimate.
Key takeaways
  • 1A 5-second AI-generated video emits about 41 gCO2e, equivalent to driving 290 metres.
  • 2A video 2 times as long needs 4 times as much energy to generate.
  • 3Measurements span a ratio of 1 to 3,000, depending on the generator and its settings.
  • 4Generating 1 video a day for a year emits 15 kgCO2e, equivalent to driving 110 km.

Generating a 5-second video with artificial intelligence (AI) emits about 41 gCO2e (grams of CO2 equivalent) with world average electricity, equivalent to driving 290 metres. According to a study published in September 2025, the most downloaded open video generator (also known as an open text-to-video model) uses 90 Wh (watt-hours) of electricity per video (in French), equivalent to 31 AI-generated images or 300 written questions.

AI-generated video

A 5-second video emits as much as 31 images or 300 questions

The area of each square is proportional to emissions, using world average electricity.

Epoch AI (2025); Luccioni, Jernite and Strubell (2024); Delavande, Pierrard and Luccioni (2025); Ember, Global Electricity Review 2026 (AI Impact Database); ADEME, Impact CO2. Projet Celsius calculation, server electricity only.

A video 2 times as long needs 4 times as much energy. Published measurements range from 0.06 to 190 gCO2e depending on the generator and its settings, a ratio of 1 to 3,000. Google, OpenAI and the Chinese video apps publish no measurement for their own services.

1How much energy does an AI-generated video use?

About 41 gCO2e for a 5-second video at 720p

The measurements come from 3 researchers at Hugging Face. In September 2025, they published the energy use of 7 open video generators (in French) (generators whose code is public), tested on a data centre graphics card. Their reference model is the most downloaded of them: it uses 90 Wh for a video of 81 frames at 720p (the resolution of an HD screen), a little over 5 seconds long. It is the second most energy-intensive of the 7, while the median model, which produces lower-resolution videos, uses 25 Wh, or 12 gCO2e.

An Nvidia server tray with 8 graphics cards, pulled out of its chassis and placed on a table
An Nvidia HGX B200 tray with 8 graphics cards, photographed in 2025. The 2025 measurements were made on a single card of the previous generation: the graphics card accounts for more than 80% of the energy of a video, with the rest going to the processor and memory. Photo by Pokiiri, Wikimedia Commons, CC BY-SA 4.0

A ratio of 1 to 3,000 depending on the generator and its settings

In the same study, the least energy-intensive generator uses 0.14 Wh, or 0.06 gCO2e, for a video of less than 2 seconds at low resolution, while the most energy-intensive uses 415 Wh, or 190 gCO2e, for 5 seconds at 720p. This ratio of 1 to 3,000 reflects the default settings (duration, resolution, number of computation passes) as much as the generators themselves. The published figures also disagree with one another: for the same model, the International Energy Agency (IEA) puts the graphics card alone at 115 Wh, while the 2025 study measures 22, and nothing explains this difference of 1 to 5.

AI-generated video

Depending on the generator and its settings, a ratio of 1 to 3,000

Emissions per video, global electricity, default settings. Each tick mark is 10 times the previous one.

0.1 g70 cm
1 g7 metres
10 g70 metres
100 g700 metres
0.06 gCO2eAnimateDiff, 1.6 s
1.7 gCO2eLTX-Video, 5 s
4.4 gCO2eCogVideoX 2B, 6 s
12 gCO2eCogVideoX 5B, 6 s
14 gCO2eolder version of a model, 5 s
24 gCO2eMochi 1, 3 s
41 gCO2eWAN 2.1 (1.3B), 5 s, the article's value
53 gCO2eIEA reference, 6 s
190 gCO2eWAN 2.1 (14B), 5 s
430 gCO2enewer version of the same model, 5 s
  • 7 open models measured in 2025
  • 3 frequently cited figures
Delavande, Pierrard and Luccioni, Video Killed the Energy Budget (2025), default settings of each model; MIT Technology Review (May 2025); IEA, Energy and AI (2025), graphics card only. Conversion: Projet Celsius calculation.

Equivalent to 31 images or 300 written questions

An AI-generated image uses 2.9 Wh on average, and a written question put to an AI assistant (also called a query) uses 0.3 Wh. The 5-second video (in French) is therefore equivalent to 31 images, or 300 questions.

The same video also uses as much electricity as 1 hour 10 minutes of streamed video, screen and network included, based on the 77 Wh per hour that the IEA calculated for 2019. Data centres account for only 5% of that hour: counting the servers alone, generating 5 seconds is equivalent to streaming 23 hours. Our article on the carbon footprint of TikTok covers video viewing.

No measurement published by Google, OpenAI or the Chinese apps

In early October 2026, there was no published measurement from Google for Veo, from OpenAI for Sora (shut down in 2026), from Kuaishou for Kling or from Runway. These values therefore describe open models, and nothing indicates whether commercial services do better or worse.

They also come from a single team, Sasha Luccioni's group at Hugging Face, which is behind the measurements picked up by the IEA and by MIT Technology Review. The environmental impact of AI also plays out at the scale of data centres, which per-video measurements do not capture.

2Why an AI-generated video uses so much energy: frames, duration and resolution

Dozens of frames calculated together

A video is a sequence of frames, from 8 to 30 per second for the generators in the 2025 study. The generator calculates the whole sequence in one go, so that objects and movements stay consistent from one frame to the next. Every additional frame therefore enters the calculation for all the others.

Plate of 12 black-and-white photographs of a mare at a gallop with a jockey on her back, with its original title card
Eadweard Muybridge's "The Horse in Motion": 12 views of a mare at a gallop, taken in Palo Alto on 19 June 1878. In 2025, the most downloaded open generator calculates 81 frames to produce a little over 5 seconds of video. Photo by Eadweard Muybridge, Library of Congress, Wikimedia Commons, public domain

A video 2 times as long, 4 times as much energy

The 2025 study measured this on the most downloaded generator: doubling the duration multiplies the energy by about 4. A video 2 times as long (nearly 11 seconds) would therefore emit 170 gCO2e instead of 41. The rule was verified up to 100 frames, or nearly 7 seconds; beyond that, the figures are extrapolations.

AI-generated video

A video 2 times as long emits 4 times as much

Set the duration and the resolution: when the duration doubles, energy use is multiplied by 4.

Resolution
Electricity
1600gCO2e1 s5.4 s10.8 sbeyond 100 frames: extrapolation41 gCO2e

5.4 s at 720p

41gCO2e

  • 290 metres by petrol or diesel car
  • 5 phone charges
  • 1 h 10 min of streamed video

5.4 seconds at 720p, or 81 frames: the video measured in the study. Going from 720p to 1080p multiplies emissions by 5.

Delavande, Pierrard and Luccioni, Video Killed the Energy Budget (2025); Ember, Global Electricity Review 2026 (AI Impact Database); ADEME (the French Agency for Ecological Transition), Impact CO2; EPA; IEA (streaming). Order of magnitude calculated by Projet Celsius, server electricity only.

From 720p to 1080p, 5 times as much energy

The same rule applies to the number of pixels. Going from 720p to 1080p (full HD) multiplies the number of pixels by 2.25 and the energy by 5, and doubling both height and width multiplies the energy by 16. The number of computation passes (the steps through which the generator refines the image) acts in proportion: dividing the number of passes by 2 divides the energy by 2. The effects of duration, resolution and passes multiply with one another.

3"1 kWh per video" and "4 litres of water": 2 estimates and no measurement

"1 kWh per Sora video": an estimate made without data from OpenAI

The figure has 2 origins. In May 2025, MIT Technology Review published measurements made with Sasha Luccioni on an open model, CogVideoX: 944 Wh for 5 seconds with the newer version, against 30 Wh with the older one. In November 2025, a blog post titled "Every Sora AI video burns 1 Kilowatt hour" estimated that a 10-second video from Sora, OpenAI's generator, uses 0.94 kWh, or 940 Wh, based on an assumption by analysts about computing time.

Neither of these 2 values is a measurement of Sora, whose energy use OpenAI has never published. Extrapolating the rule from the 2025 study beyond what has been verified gives 0.31 kWh at 720p and 1.6 kWh at 1080p for 10 seconds, and the blog post's estimate falls within that range. The service no longer exists: OpenAI announced the shutdown on 24 March 2026 and closed the app on 26 April 2026, according to its help centre.

How much water does an AI-generated video use? Less than half a litre for 5 seconds

No study measures water per video. Applying the method from our article on images, which uses 1.1 litres per kWh to cool the servers and 3.1 litres to produce the electricity, a 5-second video uses 0.1 litre of cooling water and nearly 0.4 litre in total. The November 2025 blog post puts it at a little over 4 litres for 10 seconds, using an almost identical factor. The gap comes from the energy it assumes, which is 3 times higher than the 0.31 kWh in our extrapolation (that value would give 1.3 litres). The scope of these factors is detailed in our article on water consumption by AI.

41 video a day for a year: equivalent to driving 110 km

From 15 to 450 kgCO2e a year depending on use

Generating 1 video of 5 seconds a day for a year emits 15 kgCO2e, equivalent to driving 110 km. At 30 videos a day, the free limit Sora offered in October 2025 (according to Business Insider), the total would reach 450 kgCO2e with our reference figure. That is 5.5% of the annual carbon footprint of a person in France, or a fifth of a Paris-New York return trip. At the time, the head of the service called its economics "currently completely unsustainable". There is no reliable count of videos generated worldwide.

3 ways to emit less

  • Shorten the video: dividing the duration by 2 divides the energy by 4.
  • Lower the resolution for test runs: 480p needs 5 times less energy than 720p.
  • Prepare your prompt (the request you give the generator): each abandoned attempt uses as much energy as the video you keep.

The first of these appears in our ranking of actions that reduce the impact of AI. A company that produces videos with AI can include them in its purchases of digital services, as our guide to AI in the Bilan Carbone® (the French carbon accounting method) explains. Our AI carbon footprint calculator gives a first estimate.

5Key takeaways

  • About 41 gCO2e for 5 seconds at 720p, equivalent to driving 290 metres, for the most downloaded open generator.
  • From 0.06 to 190 gCO2e depending on the generator and its settings: a ratio of 1 to 3,000.
  • A video 2 times as long needs 4 times as much energy; going from 720p to 1080p needs 5 times as much.
  • No published measurement exists for commercial services: the "1 kWh per Sora video" is an estimate.

Our articles on AI electricity consumption at world scale, on choosing a more eco-friendly AI and on whether you should boycott AI take these orders of magnitude further.

Further resources

Frequently asked questions

or: guillaume@projetcelsius.com

More articles to read

View all →
French environmental cost label for clothing: what changed on 1 October 2026
Regulation

French environmental cost label for clothing: what changed on 1 October 2026

6 min read
What is the French textile eco-score?
Who displays it, and has it become mandatory?
Regulation

What is the French textile eco-score? Who displays it, and has it become mandatory?

12 min read
Removable batteries and battery passports: who is affected in 2027?
Regulation

Removable batteries and battery passports: who is affected in 2027?

14 min read

A senior consultant reads every message

Tell us where you stand and what is pressing, and we will reply within 24 working hours with a first view on how to proceed.

Write to a consultant →
Our toolkit

A full range of free assessment tools

Cost estimator, obligation and eligibility checkers, footprint calculator.

Estimate costs and impacts
Ref. 2026·BC·0184REPORT · FINANCIAL YEAR 2026Bilan Carbone®complete 1·2·3Presentation to the Executive Committee · 2027 action planCONTENTS01Executive summaryp. 402Scope and methodologyp. 1203Emissions by scopep. 24042027 action planp. 5605Methodological appendicesp. 78VOLUME84 pages · ConfidentialSTANDARDISO 14064 · GHG ProtocolQUOTATION · BC.2026.0184Issued 05·05·2026 · Valid for 30 daysCelsius consultancy · Paris 3rd arrondissementINDICATIVE COST ESTIMATEYour companyMid-sized company · 180 employees · CSRD 2027SERVICEBilan Carbone®Scope 1·2·3 · 13 weeksBPIFRANCE FUNDING · - 60%INDICATIVE RANGE EXCL. VAT€16,800to €22,400PROJECT DETAILS01 · SCOPING3 weeks~€4,20002 · MODELLING8 weeks~€9,80003 · PRESENTATION2 weeks~€4,800Indicative estimate · Refined after scopingSIRET 891 234 567 00012

Bilan Carbone® cost estimator

Your price range in 1 minute, based on 2026 market rates and public funding.

1 minStart
Estimate costs and impacts
AI FOOTPRINT · MONTHLY USEYOUR TOTAL FOOTPRINT4.2 kgCO₂eq for 1,000 typical prompts · May 2026MODEL COMPARISON · 1,000 TYPICAL PROMPTSGPT-4oOpenAI6.8 kgClaude OpusAnthropic2.1 kgGemini ProGoogle4.5 kgMistral LargeMistral AI · FR1.6 kg

AI carbon footprint calculator

The climate impact of your AI queries, by model and task. Ecologits methodology.

3 minStart
Check an obligation
DIAGDÉCARBON'ACTIONACCELERATE THETRANSFORMATIONOF YOUR COMPANYIN ENERGY ANDTHE ENVIRONMENTOFFICIAL SCHEME · BPIFRANCE × ADEMECHECKER · ELIGIBILITYELIGIBLEYou meet the 3 Diag Décarbon'Action criteria.YOUR FUNDING€10,000 excl. VAT€6,000excl. VATAmount payable by you after the Bpifrance subsidyFUNDING BREAKDOWNBPIFRANCE 40%YOU 60%€4,000€6,000SCHEME SUBSIDISED BY

Diag Décarbon'Action eligibility

Check in 30 seconds whether your business is eligible for Bpifrance funding covering 40% of a Bilan Carbone® assessment.

30 secStart
Check an obligation
FRENCH REPUBLICMinistry for the Ecological TransitionLEGAL REQUIREMENTGreenhouse GasEmissions AssessmentArticle L. 229-25 of the French Environmental CodeDecree 2022-982 · published 1 July 2022FREQUENCY · 4 YEARSFILING · ADEMECHECKER · BEGES REQUIREMENTREQUIRED720 employees · BEGES requirementPOTENTIAL PENALTY€50,000maximum fine for failing to complete itart. L. 229-25OFFICIAL REFERENCE

BEGES checker

Does the BEGES requirement apply to you? An immediate answer, with the deadline and penalty.

30 secStart
Check an obligation
9:42SCANSCANNING DPP...DPP IDENTIFIEDOrganic cotton T-shirtSKU TX-CB-220 · Batch L-26-04781ESPR COMPLIANT42 DATA POINTS · 5 SECTIONSORIGINIndia · GOTSOrganic cottonFOOTPRINT5.2 kg CO₂eqLCA ISO 14040RECYCLABILITY85%Cotton recycling schemeREPAIRABILITY7.5 / 10Accessible partsEND OF LIFETextile recycling scheme (TLC) · Bordeaux/FRRefashion-approved recovery operatorEU · ESPR 2024/178112.05.2026

DPP checker

Is your product covered by the Digital Product Passport?

1 minStart
Check an obligation
COSTENVIRONMENTAL386POINTS257PER 100 GOfficial methodEcobalyse, v7.0.0CHECKER · ENVIRONMENTAL LABELLINGELIGIBLETextiles · voluntary labellingENVIRONMENTAL COST386 ptsT-shirt 150 g · 257 pts per 100 gExample: Ministry for the Ecological Transition FAQ

Textile environmental labelling 2026

Voluntary labelling, but third parties can publish it from October 2026: where do you stand?

1 minStart