Answers before you ask
Delivery, refunds, compatibility, licensing and updates. Anything else — write to us, a human replies within one business day.
Instantly. The moment your order is confirmed you receive an email with a download link to a organised workspace: a searchable PDF, a Notion database, and plain Markdown files you can drop into any tool. The whole delivery takes under a minute — no accounts, no waiting.
Every pack lists its compatible models on the card. Most prompts are model-agnostic and tested across ChatGPT (GPT-4o and later), Claude (Sonnet and Opus), and Gemini. Midjourney packs are written for Midjourney v6+ syntax. Where a model needs a variant, we include it in the pack.
When we improve a prompt, add new prompts to a pack you own, or adapt packs for new model releases, you get the new version free — forever, via the same download link. Since launch we've shipped 14 pack revisions; owners of the original versions received every one.
30 days, no questions asked. If a pack doesn't save you time or improve your output, email us and we refund the full amount within 3 business days. We can offer this because our refund rate is under 2%.
Yes. A purchase includes a commercial licence: use the prompts in your work, for clients, and inside your company's tools. The only restriction is redistribution — you can't resell or republish the prompt packs themselves, in whole or in part.
Packs are one-time purchases: you buy once, you own them forever, including updates. Pro is a subscription that unlocks the entire library — every current pack and every new monthly drop — for as long as you stay subscribed. Pro members also get the prompt-chaining workflows that connect packs together.
The prompts are written in English, but all major models follow them well and respond in whatever language you request. We include a one-line 'respond in [language]' suffix pattern in every pack for localised output.
Each prompt goes through our three-stage review: it must produce usable output across at least three frontier models, survive adversarial inputs (vague briefs, missing context), and outperform a naive one-line prompt by a measurable margin in blind scoring by our team. Roughly 40% of candidate prompts don't make the cut.