A: Many projects have a fixed cost and a variable cost. This means that, even if you ask for a 3K word article from Youbooks (you can't, the form will stop you!), Yobooks would run the same workflow that makes absolutley no sense for such a length. The workflow will run for 1 hour, and produce something that you could have done in ten minutes iterating in ChatGPT. It's like trying build a sand castle with a excavator. We still need to pay for the excavator, and so we need to tarnsfer the costs.

By using a flat fee, we reflect the true economics that we have to deal with, on the price. This is good for us (less risk) and the customer for two reasons.
1) By taking less risk, we can offer lower prices
2) By making smaller books more expensive, we guide the user towards longer books, which is what Youbooks is built for

You might have noticed that 1 credit is not exactly 1 word, it is actually 1.1 words. So, at 100K credits, you get around 100K words (flat fee included). It just so happens that non-fiction books are around 100K words. Also, the editing sweet spot is also there (this is more technical and has to do with our workflow; if you care, I can explain).

So, in summary, the 9K flat fee is necessary and good. The credits are not used for a specific thing. They are amortized over the entire workflow. For example, when you ask for a 15K word book with a humanized model, we still need to keep the model in the GPU in memory for a couple of hours, vs 3 hours for 30K word project. If you plot this on a chart, the "line" crosses the y-axis on a psoitive value, which basically means, there is a fixed cost. This is only an example, the same principle applies to the entire workflow.

---

On the humanized models - we basically use your samples to train a model to write like that. How well it works depends on the quality and quanity of the samples. There is a sample book on the home page where we applied the Benjamin Franklin model (this is a factory model available to everybody). Please take a look and judge for yourself. Our metrics (loss) suggest that the model writes like Franlkin, but it is subjective.

Youbooks has its own model training pipeline. We train a small model for every human model you create. We store it, and deploy it on our GPUs.

---

If the output is very bad or something went completley wrong like mixed language, we refund the project. If the user does not think the project is good, we take a look and run an eval with ChatGPT Pro which gives a rating. If the rating is below 5, we refund it. If it is 5-6, we usually don't and explain to the user that this is the state-of-the-art at this point in book generation. Other AI book generator rate around 2-3. Self-published non-fiction on Amazon is a 5-7, and published non-fiction is 7-9. You can get there yourself by using ChatGPT Pro on a Youbooks produced manuscript, and iterating on the feedback using your judgement.

Finally, if there is one thing I can promise, this is that your book will not be perfect. It will have "bugs" and issues. If you want to publish non-fiction under your name, using Youbooks, you will definitely want to edit. If you want to publish non-fiction under ghost-writer names, you should still edit (with AI or without).