Moana is an AI-created project. Every image, every song, every video, every text you see here was generated using cloud-based AI tools. That takes energy — and we believe you deserve to know how much.
This page documents our estimated monthly energy footprint. We say "estimated" because none of our AI providers publish exact per-request energy data. Instead, we rely on the best available scientific research to calculate approximate figures. We update this page quarterly and revise our methodology whenever better data becomes available.
Full transparency: we cannot measure what the companies themselves won't disclose. But we can count what we produce, apply peer-reviewed estimates, and show you the math.
The Tools Behind Moana
All of Moana's content is produced externally — nothing runs on local hardware beyond a standard laptop used for coordination. Here are the AI services we use:
| Tool | Purpose | Provider |
|---|---|---|
| Genspark (LLMs) | Text creation, research, quality control, website development | Genspark |
| Gemini | Social media comment management | |
| NanoBanana2 | Image generation | Genspark |
| Grok Imagen | B-Roll video generation | xAI |
| ElevenLabs | Voice synthesis (Moana's voice) | ElevenLabs |
| Suno AI | Music generation | Suno |
| Veo Video | A-Roll video with lip sync | |
| CapCut | Video editing & export | ByteDance |
Estimated Monthly Energy Footprint
The following table shows our estimated energy consumption for a typical production month. Quantities are based on our actual output; energy-per-unit values are drawn from scientific studies (sources listed below).
| Activity | Energy per Unit (est.) | Typical Quantity / Month | Estimated Total |
|---|---|---|---|
| LLM text prompts (Genspark, Gemini) | 0.3 Wh per prompt | 500 prompts | ~150 Wh |
| Image generation (NanoBanana2, Grok Imagen stills) | 2.5 Wh per image | 60 images | ~150 Wh |
| Voice synthesis (ElevenLabs) | 3 Wh per minute of audio | 30 minutes | ~90 Wh |
| Music generation (Suno AI) | 40 Wh per song | 4 songs | ~160 Wh |
| Video generation (Veo, Grok video) | 90 Wh per short clip | 8 clips | ~720 Wh |
| Video editing & export (CapCut) | 5 Wh per export | 8 exports | ~40 Wh |
| Total | ~1,310 Wh ≈ 1.3 kWh | ||
What does 1.3 kWh mean?
To put this number into perspective:
- A refrigerator uses about 1.3 kWh in 1–2 days.
- An average German household consumes roughly 125 kWh per month — Moana's AI footprint equals about 1% of that.
- A single economy-class flight from Berlin to Tahiti burns approximately 12,000 kWh of energy per passenger.
- Charging a laptop for 20 days uses about 1.3 kWh.
The biggest energy consumer in our workflow is video generation, which accounts for more than half of our total footprint. Text prompts, by contrast, are almost negligible.
Methodology & Sources
Our estimates are based on the following peer-reviewed and publicly available research:
LLM prompts: Google reported 0.24 Wh per median Gemini query in August 2025 (source: MIT Technology Review). Epoch AI estimated GPT-4o at approximately 0.3 Wh per query (source: Epoch AI). We use 0.3 Wh as a conservative middle estimate.
Image generation: The Hugging Face study by Sasha Luccioni et al. found approximately 2.9 Wh per image for diffusion models (source: CNET). Other estimates range from 1.7 to 20 Wh depending on resolution and model. We use 2.5 Wh as a mid-range estimate.
Voice synthesis: Hypergrid Business reported approximately 0.5 Wh per voice query for AI chatbot voice responses (source: Hypergrid Business). For longer-form voice generation (full paragraphs, narration), we estimate ~3 Wh per minute of generated audio based on extrapolation.
Music generation: The ArXiv paper "Diffused Responsibility: Analyzing the Energy Consumption of Generative Text-to-Audio Diffusion Models" (2025) measured 1–10 Wh for 10-second audio clips depending on model and settings (source: ArXiv). For a full 3-minute song with multiple generation attempts, we estimate ~40 Wh per finished track.
Video generation: The Hugging Face study found approximately 90 Wh per AI-generated video clip of 5–10 seconds. WSJ testing showed a range of 20–110 Wh per clip (source: WSJ, June 2025). We use 90 Wh per clip as our estimate.
Video editing (CapCut): No peer-reviewed data available for cloud-based editing tools. Our estimate of ~5 Wh per export is based on general cloud rendering benchmarks and is likely conservative.
Honesty Note
These numbers are estimates, not measurements. None of our AI providers — not Google, not ElevenLabs, not Suno, not xAI — publish per-request energy consumption data. We rely on third-party research that itself involves assumptions about hardware, data center efficiency, and model architecture.
We publish this page not because we have perfect answers, but because we believe the question matters. If our providers begin disclosing actual figures, we will update immediately.
This page is reviewed and updated quarterly.
Last update: Q1 2026 (March 2026)
Next scheduled update: Q2 2026 (June 2026)