On sustainability at Bolkra. Last updated 21 May 2026.
Commercial image production has a carbon footprint, and the biggest line item in that footprint is not what most people assume. It is not the lights. It is not the studio. It is not the models or the wardrobe. It is the travel. Flying crews, shipping samples, freighting sets and props between cities and continents. The work of getting people and physical things to the place where the image is made.
Bolkra removes that line item from the equation. Generation happens in a data centre. Nothing flies. Nothing ships. Nothing is built only to be struck.
This page sets out what we know, what we do not, and what we will tell you when we have measured it ourselves.
60%
Share of typical advertising shoot emissions from travel and transport, in the most recent industry-wide measurement.
Source: AdGreen 2023 Annual Review, 1,424 projects measured.
6.2 tCO₂e
Average carbon footprint of a single advertising shoot in the UK industry's most current dataset. For productions over £50,000 per shoot day, the average rises to 13.9 tCO₂e.
Source: AdGreen 2023 Annual Review.
1 per season
Typical number of times a single product is reshot in a season, for colourway variants, regional adaptations, or seasonal restyling. Each reshoot multiplies the footprint linearly.
Source: industry practitioner experience. No peer-reviewed figure exists for this number, and we have flagged it as such in Section 7.
The largest avoidable emissions in fashion sit upstream of any shoot, in the garments that get manufactured but never sold. Industry estimates put unsold stock between 10 and 40 percent of annual production. From 19 July 2026, the EU bans large fashion businesses from destroying it. Bolkra-generated campaign imagery lets brands test demand before manufacturing committing volume. The cleanest garment is the one you did not need to make.
A typical advertising shoot in 2023 emitted 6.2 tCO₂e, with 60 percent of that footprint coming from travel and transport alone. For larger productions, flights accounted for 49 percent of all emissions on their own. Bolkra removes the line item entirely. Generation happens in a data centre. The crew does not fly because there is no crew to fly.
Most e-commerce imagery is produced in speculative batches: hundreds of products, dozens of variants, every angle, every colourway, every regional adaptation. Most of it never gets used in market. With Bolkra, imagery is generated at the moment of need, in the volume actually required. The marginal cost of one more variation is small enough that the workflow changes.
A typical product is reshot once per season for variants, localisation, and creative direction shifts. Each reshoot multiplies the physical footprint linearly. With Bolkra, variants are a small marginal cost on a fixed base, not a new shoot. The flights are not repeated because they are not taken.
Most fashion and product shoots require physical samples to be manufactured, shipped, and often discarded after the shoot. Apparel reverse logistics alone is responsible for tens of thousands of tonnes of CO₂ annually at large retailers. Bolkra works from existing product references and trained models. The samples that would otherwise have been made for photography are not made.
Generating an image with AI has a real energy cost, a real water cost, and a real carbon cost. They are smaller than the alternatives by orders of magnitude, on the right hardware on a clean grid. They are not zero. Section 6 sets out the honest accounting on this side of the ledger, with peer-reviewed numbers and a commitment to publish our own once we have measured them.
The EU Ecodesign for Sustainable Products Regulation comes into force on 19 July 2026. From that date, large fashion businesses (those with 250 or more employees and turnover above 50 million euros) are prohibited from destroying unsold clothing, accessories and footwear. From 2030 the ban extends to medium-sized companies. Mandatory annual disclosure of the volume and reasons for discarded unsold products applies under Implementing Regulation (EU) 2026/2, which begins reporting from 2 March 2027.
The European Environment Agency estimates 4 to 9 percent of unsold textiles placed on the EU market are currently destroyed before ever being worn, generating around 5.6 million tonnes of CO₂e annually, close to Sweden's net national emissions.
Bolkra is not a compliance product. It is a creative production tool. But for businesses producing for the EU market, the regulatory environment is now actively rewarding any creative workflow that reduces speculative overproduction. The economics and the law are starting to point in the same direction.
We are not going to pretend AI image generation is carbon-neutral. That would be the wrong answer to a real question and it would not survive scrutiny by anyone who actually thinks about this.
What we can do is tell you the honest numbers as the peer-reviewed literature reports them, and tell you what we do and do not know about how those numbers apply to Bolkra specifically.
The most rigorous current measurement of AI image generation comes from Luccioni, Jernite and Strubell at the ACM FAccT 2024 conference. They tested 88 image generation models on standardised hardware and grid mixes. They found a median energy cost of 1.35 watt-hours per image, with the most carbon-intensive model in their study emitting approximately 1.6 grams of CO₂e per image on a typical US grid. The most efficient models in the same study sat below 0.1 grams per image. The two-order-of-magnitude range is real and depends on model, hardware, and grid.
Bolkra generates images through cloud-hosted inference. The carbon footprint of that step depends on which data centre region runs the job, how efficiently the facility operates, and how clean the local grid is. Leading hyperscale facilities report Power Usage Effectiveness below 1.2; the Uptime Institute puts the industry average at 1.56. Those structural differences matter, but they are not the same as measuring Bolkra's own emissions.
Some providers publish carbon and water figures for text-only AI prompts. Equivalent figures for image generation are not yet available from our stack in a form we would trust. We will not extrapolate from text to image, because the numbers would be wrong by an order of magnitude in either direction.
So here is what we are committing to. Once Bolkra has 30 consecutive days of metered usage data from our inference infrastructure, we will publish a per-image figure for Bolkra specifically, with the methodology referenced to the GHG Protocol and both location-based and market-based Scope 2 numbers disclosed. We will update it as the inference fleet or regional grid mix changes. Until then, we are reporting what the peer-reviewed literature says about AI image generation in general, and not pretending to have measured what we have not.
For the record, on the cleanest current numbers, one mid-sized advertising shoot at the AdGreen 2023 industry average is the carbon equivalent of somewhere between 3.9 million and 200 million AI-generated images, depending on the hardware and grid the inference runs on. The range exists because the comparison is genuinely sensitive to where the data centre is and what silicon is in it. We are not going to publish the high end of that range as a marketing headline because doing so on a coal-heavy grid would be misleading. The honest claim is narrower: Bolkra replaces a physical activity whose carbon footprint is measured in tonnes with a digital activity whose carbon footprint is measured in grams. The orders of magnitude are not in dispute. The exact multiplier is.
A sustainability page that pretends to know everything is not telling the truth. Here are the questions we cannot answer rigorously today. We will answer them as we and the industry collect better data.
We can quote the industry literature on AI image generation in general, but we have not yet measured Bolkra's own metered inference usage rigorously enough to publish a single defensible number. We will, once we have 30 days of metered data, using the GHG Protocol methodology.
Industry practitioners often cite roughly one reshoot per product per season for fast-moving fashion e-commerce. No peer-reviewed study has measured this rigorously across the industry. We have flagged it as practitioner experience, not measured fact.
Documented as a standard workflow in industry operational guides. Not quantified globally. The carbon saving from eliminating it is real but we cannot put a number on it at industry scale.
No standardised lifecycle assessment exists for the land footprint of an individual unsold garment. The water and carbon footprints are better documented; land is the gap.
This is the most consequential claim a sustainability page could make about Bolkra, and it is the one we are least able to support with a number. The mechanism is real, demonstrated by Inditex's near-shore demand-led production model. The peer-reviewed pilots quantifying the reduction achievable through AI-based demand signal do not exist yet.
The published range is 2 to 8 percent of global emissions, depending on study scope and methodology. Most credible mid-range estimates fall between 3 and 4 percent. We do not have an answer that is meaningfully tighter than this range.
The sustainability claims a vendor makes are only worth the measurements they are based on. We commit to the following on a rolling basis.
If you are an agency or brand-side sustainability lead and you want to talk about how Bolkra usage fits into your own carbon accounting, we would rather have that conversation properly than send you a marketing page.