Split a long novel into translation batches: two points the app has already locked down, and one job you do at every batch boundary
Tri Thu知秋
On this page
- Where does a 300-chapter novel actually break between batches?
- What does locking one dictionary at the start of a batch mean, and can the AI hold all 300 chapters in mind?
- Back the next day, how do you know where the last batch stopped?
- How many credits does one batch cost?
- At the batch boundary, what do you have to review in the name table?
One batch takes at most 50 chapters or 200,000 source characters, so a 300-chapter novel has to run as several batches, at least six if you split evenly by the chapter cap. Between batches, the system already keeps two things in place: the dictionary is locked as one set the moment a batch starts and the chapters of that batch keep working from it, and "Continue translating" gathers exactly the untranslated chapters until the cap is reached. The job that remains is yours: review the name table and the forms of address at the batch boundary, then start the next batch.

| Boundary | What the app holds | What you hold |
|---|---|---|
| Before the first batch | One dictionary and its settings are locked the moment the batch starts | Check whether the name table already covers this batch's characters; the dictionary cap is 500 entries |
| While a batch is running | Saved chapters do not disappear when you close the tab; a mid-run error state is recorded so it can be handled when you come back | No intervention needed in the middle of a batch |
| At the end of a batch | The batch is cut exactly when it reaches 50 chapters or 200,000 characters, never silently over the line | Review the name table and the forms of address before starting the next batch |
| Starting the next batch | Continue translating gathers exactly the untranslated chapters, even when they do not form one stretch | Fix the name table first, then press run |
Where does a 300-chapter novel actually break between batches?
The break usually looks like this: today you finish batch one, a few days later you no longer remember which chapter the old batch stopped at, and a hand count either repeats chapters you already translated or skips a whole stretch. Worse is the wording: a later batch changes how characters address each other compared with the batch before, and readers spot it the moment their eyes touch the page.
Look closely and the break sits at only three points. Two of them are already locked down by the system: the dictionary is locked as one set when a batch starts and its chapters keep working from that set; and "Continue translating" gathers exactly the untranslated chapters, so you do not have to remember how far you got. The third point is human work: reviewing the name table at the batch boundary. This article follows exactly those three points.
Before going point by point, one thing needs to be said plainly: the app does not schedule the whole novel's split for you. The 50-chapter or 200,000-character cap only cuts when a running batch reaches it; the order of the batches and the moment you review the name table are your decisions. Splitting into batches is step 4 of the five-step workflow for translating a long novel; the whole workflow is in How to translate a whole web novel with AI: a 5-step workflow.
What does locking one dictionary at the start of a batch mean, and can the AI hold all 300 chapters in mind?
When a batch starts, the system locks one dictionary and one set of settings for that batch. While the batch keeps running normally, including a retried chapter or a short pause before resuming, its chapters keep working from that locked set. The one thing that can make the rest of the batch see a different table is editing the name table mid-batch: if the set has to be captured again midway, the system picks up the table as it stands at that moment. So settle the name table before you press translate, and leave it alone mid-batch. How the characters are addressed is still the part you check in the translation.
Across batches, do not picture an AI that remembers the full context of all 300 chapters, because the translation context runs batch by batch. What stitches the batches together is the dictionary: the name table and the forms of address you have reviewed, not a model's memory still holding the whole novel intact.
That name table has a cap of 500 entries, so do not pour the novel's whole world into one table. Inserting entries by the main character groups of the batch about to run is enough; a later batch's side characters can be added when you review that batch's boundary.
Back the next day, how do you know where the last batch stopped?
The "Continue translating" button answers exactly this question: it gathers the untranslated chapters from the mark you choose until the batch cap is reached. Already translated chapters are skipped, so a batch's chapter range can perfectly well be non-contiguous, for example patching a few chapters left empty in the middle of the novel. Keeping track by hand on paper goes wrong even more easily: one mistaken line and the next batch repeats chapters or skips ahead.
When you deliberately want to redo a contiguous range, including chapters already translated, use the "Choose a translation range" mode from one chapter to another. The two modes split the work cleanly: Continue translating to carry the novel forward, Choose a translation range to retranslate exactly the stretch you point at.
If you close the tab mid-translation, the chapters already saved stay where they are. If an error hits midway, whether credits come back depends on the kind of error, spelled out in the cost and refund article. A single unusually long chapter that passes even the 200,000-character cap can still be translated, on the condition that the batch contains that chapter alone.
If your novel lives in an EPUB file, the full five-step walkthrough from uploading the file to exporting the translation is in How to translate an EPUB novel into English and export a file you can read. Why a general chat window usually cannot translate a multi-chapter EPUB in one pass without hitting output limits is covered in Can ChatGPT read and translate a full EPUB novel? Output limits and how to keep the table of contents.
How many credits does one batch cost?
Credits count source characters, spaces included; a normal batch is limited to 200,000 characters, while a single over-cap chapter can run as a batch of its own. The credit limit for each batch is therefore known before you press; converting credits into money, package prices and the cases where credits come back live separately in Novel translation cost with AI: what you pay, and do you lose credits if it fails midway?
A single upload accepts up to 5 million source characters, so the batch-splitting boundary sits at the translation step, not at the file upload step. If you are starting with the 200,000-credit trial on your first batch, see Free AI novel translation trial: how the 200,000 source-character credit works. When you need to top up for later batches via PayOS or Creem, see Topping up novel translation credits: one-time packages, international cards via Creem, and credit expiry.
At the batch boundary, what do you have to review in the name table?
This is the point the app does not do for you. The batch boundary is when you open the name table and go over the characters of the batch just finished and the batch about to run: main names, nicknames, down to how each pair of characters addresses each other. The new batch's dictionary follows exactly the table as it stands, even when that table is still missing entries or still wrong, so fixing it before you press is always cheaper than fixing it afterwards.
The review belongs exactly at the boundary: the old batch has already run its course, and the new batch has not started, so there is still time to change anything. How to build the name table from scratch, choose main names and gather nicknames is in Build a character name table before translating a novel with AI.
If you find a wrong form of address that has already spread through earlier batches, do not retranslate the whole novel: picking exactly the passages that still use the old wording and retranslating them is enough; the how-to is in Changed a name or a term: how to retranslate passages that still use the old wording.