How to Budget AI Drafting for a Full Novel

You do not need a personal API key to draft a novel here. Free, Pro, and Ultra all use included Kestrel pools. The budget problem is the same as it ever was: regenerating whole chapters because one paragraph failed.

Rough math

An 80,000-word novel might be 35–45 chapters at 1,800–2,500 words each. Each chapter generation includes bible context, a prior summary, the outline chunk, and the output. Outline, writing instructions, polish, retries, and edits all spend the same pool — not output-only capacity.

On Free, plan around a full outline plus a paced handful of new chapters per day (edits do not use a new-chapter slot). On Pro and Ultra, a monthly pool replaces the daily gate; Ultra is a larger pool of the same powerful model family as Pro.

Task-based model choice

• Outlines and beat sheets — Free AI is enough for structure; save Pro/Ultra for prose you will keep
• Final prose — Pro or Ultra when voice matters
• Line polish and targeted edits — scope the prompt to the broken scene instead of regenerating the chapter

Do not regenerate an entire chapter because one paragraph failed—rewrite the scene in the diff editor.

Allowance-control habits

1. Freeze the bible length; trim instead of appending
2. Accept "good enough" first drafts; revise with diffs
3. Keep plans finishable (Free-friendly lengths) so you are not mid-book when the daily reset hits
4. Upgrade only when you want stronger prose and a monthly pool, not because Free is "missing a key"

Compare models on our AI model comparison page. Story Wizard walks the outline → chapter workflow with included Free AI — no setup.

What actually spends the pool

Authors budget as if only “new chapter” words count. On Entangled Text, outline generation, writing instructions, chapter context, polish, retries, and typical edits spend the same Pro/Ultra pool. A 2,000-word chapter that you regenerate three times is not 2,000 words of allowance — it is closer to a small disaster. Scope the prompt: “rewrite the last 400 words; keep the betrayal beat.”

Free is paced (outline plus a handful of new chapters per day; edits do not use the new-chapter slot). That is a feature if you revise as you go. It is a blocker if you want a weekend binge. Do not “wait for the daily reset” by regenerating yesterday’s chapter until it is perfect. Accept a draftable chapter, lock what you like, move on.

A finishable 80k plan

1. Outline first, once. Fix plot holes as two-sentence beats. Regenerating the whole outline weekly is how you never write chapter four.
2. Sample chapter, then stop and edit voice. If you hate the sample, change bible/voice notes — do not generate chapters 2–20 in the same voice.
3. Draft forward. Do not polish chapter one to publication while chapter twenty is still a beat. You will rewrite chapter one when the ending teaches you what the book is.
4. Keep a regeneration log: date, chapter, why. If the “why” is “vibes,” do not spend the pool.
5. Upgrade to Pro when Free’s daily gate is the bottleneck and you have an outline you trust. Ultra is for a larger pool of the same powerful family, not a different product.

Top-ups ($5) on Pro/Ultra are for a mid-cycle squeeze, not a substitute for scoped prompts. Unused paid-pool words do not become a reason to regenerate for fun. See pricing for what cancellation does: you return to Free’s pace, you do not keep a leftover Pro bucket.

Context is not free either

A 4,000-word bible plus three prior chapters pasted into every job is a tax. Trim the bible; summarize prior chapters instead of pasting them whole when the studio already carries a running summary. If you are on Free, wasting the daily new-chapter slot on a chapter you will delete because the bible was stale is the expensive mistake — not “Free is too slow.”

Write a one-line goal for the chapter before you click generate. If you cannot write the line, you are not ready to spend the pool. That habit saves more allowance than any model switch.