TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Ethics & Provenance · AGE 17—20 · FACTUAL · EVOLVING
Generation is not free
Every generated image draws electricity somewhere, and the discipline is a tighter brief.
The idea
Every image you generate is computed somewhere that draws electricity: a data centre, often with cooling water, not always; a small model on your own machine, electricity alone. Every API call has a price too, or a cost the provider carries. The energy per image is argued over and changes with the model, the hardware and the task (a text reply and a minute of video are nowhere near each other), so this card gives you the scale, not a number. It is not asking for guilt. It asks for an architect's discipline: price a detail before drawing five hundred.
Why it matters
You already carry climate responsibility as a designer. This is the same responsibility, reaching into a building you will never see.
See it in the studio
A student generates fifty images for one context view and rejects forty-nine. Who paid for them? Not the model. A brief that could not say what it wanted paid for them, in money, in energy, and in the hour the student lost.
Watch for this
Two opposite mistakes. "My one prompt is nothing" is true on its own, and false at the scale every prompt is now run. And "AI is destroying the planet", said without a number, is a feeling, not a measurement. There is no universal "one prompt = so many millilitres" figure: anyone who gives you one without a facility, a model and a date is guessing.
Try it
Look up the per-image or per-token price of one tool you use and estimate what your last session cost in money. Then write the tighter brief that would have reached the same result in a third of the tries.
Prove it
Estimate the rough cost of your last generation session in energy and money, stating which parts are known and which are guessed, and show how a tighter brief would have reached the same result for less.
How it works
Training a frontier model is a large cost concentrated in model-building runs, repeated each time a model is retrained; everyday use, called inference, is small per call and enormous in total. The IEA's 2025 Energy and AI report is the best public picture of the scale, and it will be revised — which is why this card is EVOLVING. The Trained rarely, used constantly card on this map separates the two bills. The Six-layer brief on the Brief strand is the practical tool that lowers the inference bill by asking properly the first time.
What this idea builds on
What this idea opens up
Sources
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.