Frequently Asked Questions
Seven questions about pricing accuracy, missing categories, and how to use this calculator for production budgets.
Frequently asked questions
Why does the same Shorts cost $12 on one model and $30 on another?
Because video generation alone swings 2.5x across mainstream models in 2026. Kling 2 charges $0.20 per second of 1080p output, Runway Gen-3 charges $0.40, and Veo 2 charges $0.50. For a 60-second Short that single line item moves the total by $12 to $30 before you have even touched the script, voice, or thumbnail stages. The calculator surfaces this swing as the "best mix" panel below the total.
Are these prices real list prices or estimates?
They are public list prices as of 2026-05, taken from each vendor's pricing page or amortised from their subscription tiers when a per-unit rate is not published. ElevenLabs and Midjourney are amortised; OpenAI, Anthropic, Google, Runway, and Kling publish per-unit rates directly. The full snapshot is committed at /prices.json under CC0 so you can audit every number.
How accurate is the script token estimate?
Within roughly 10 to 15 percent for English and 15 to 25 percent for Korean. We use 2.2 spoken words per second and 1.4 BPE tokens per word as constants, which match GPT-4o's tokenizer well for English prose. Korean tends to use more tokens per character on legacy tokenizers, so total cost on Claude or Llama 3 is biased low. For production budgets, override the LLM stage with your own measured token count from a sample script.
Does this site upload my prompts or content to a server?
No. Pricing data is bundled into the static build and every calculation runs entirely in your browser tab. Your scenario picks are saved to localStorage, and the share link encodes state in a base64 URL token. There is no backend, no analytics of your inputs, and no account system — paste budgets containing client names or unreleased campaigns without exposure.
What is missing — fine-tuning costs, image-to-video upscaling, GPU rental?
Fine-tuning, custom-voice cloning, upscaling, b-roll editing, music licensing, and human review are intentionally out of scope. They are usually one-time or per-project costs that do not scale per piece, and including them would make the per-piece number misleading. Add them as a separate fixed overhead in your monthly budget; the calculator gives you the variable cost.
Why does monthly volume only multiply per-piece cost without volume discount?
Most frontier AI APIs in 2026 do not offer transparent volume tiers below enterprise contracts (typically >$5,000/mo committed). For solo creators and small studios producing 5 to 50 pieces a month, list prices are what you actually pay. If you have a real enterprise agreement, override the per-stage model with your contracted rate by editing /prices.json locally and self-hosting.
Can I get the raw price dataset for my own internal tools?
Yes. The full price catalogue is published at https://ai-content-cost.bal.pe.kr/prices.json under CC0 (public domain dedication). Fork it, mirror it, or replace it with your own contracted rates — no attribution required. It is also crawl-friendly so LLMs can cite specific per-unit numbers in their answers.