1667write in the terminal
menu
forkinstall

Migration guide

Move from NovelAI

Read the published guide.

This guide shows how to import NovelAI .story, .scenario, .lorebook, and .preset files into 1667. It explains supported transfers, import limits, data storage, and model access.

The concept map

NovelAI 1667 What changes
Memory An always Fact Import gives the Fact the memory tag
Lorebook Entry A Fact Entry keys and activation become Fact settings
Author’s Note Author’s Note The text imports; the position resets to the default depth
Retry Take A retry can become one take when it adds fresh prose without changing earlier prose
Sampler Preset Generation Profile Supported sampling values become Profile settings
.story file Story One file becomes one story
.scenario file Story The prompt becomes the first story parts

A Fact is one note that 1667 sends with a provider request. An always Fact is in each continuation request and rewrite request. A keyed Fact enters the request when the story text matches one of its keys. 1667 ignores letter case for literal keys. Regex keys are case-sensitive unless they use the i flag.

Import turns each Lorebook Entry into one Fact through the Entry Mapping. Its text becomes the Fact text, and its keys become the Fact keys. An always-on entry arrives as an always Fact. A disabled entry does not import.

A keyed Fact can use literal or restricted regex keys. It can also use secondary keys with AND or NOT logic. Scan depth sets how many recent story parts 1667 searches. Chain activation lets the text of an active Fact activate other Facts.

The Author’s Note keeps its name. 1667 sends it near the end of each continuation request and prompted retake request. Import reads the note text. It does not read the NovelAI placement. The imported note lands at the default Author’s Note depth: immediately before the last story part. Change the depth in the Author’s Note editor.

A NovelAI retry maps to a take. A take is one alternative version of a story part. 1667 keeps each imported take, and the mass map shows all of them. Import reads the retry history from a .story file. The retry you had open in NovelAI becomes the selected take for its story part.

Import does not read NovelAI’s own generation settings. The Fidelity Report states this omission.

Import keeps a retry only when it adds fresh prose after an existing story part or an earlier retry. Import drops a retry that edits or removes prose already in the story line. It states the dropped count in the Fidelity Report. A dropped retry never changes the selected story line, which import always reads in full.

Import reads retry history from the current NovelAI Document format. It also reads the datablocks history from the legacy story format. A legacy retry must start after a complete imported story part. Import drops a legacy retry that starts inside one story part. This rule prevents the import from changing or omitting shared prose.

Import can omit the complete retry history if replay work reaches its safety limit. The Fidelity Report states retry history omitted: replay work limit reached.

Before you start

Export each file from NovelAI:

  • In the Story tab, select Export Story, then To File for a .story file or As Scenario for a .scenario file.
  • In the Lorebook, select the export button for a .lorebook file.
  • In the Config Preset settings, select the export button for a .preset file.

Then make a 1667 project:

1667 init

Import a story or a scenario

Give each .story or .scenario file to 1667 import:

1667 import alderaan.story

The command makes one new story for each file. It does not write to the file that it reads. It prints one line for each file:

alderaan.story: imported "Alderaan" (312 parts, 24 facts) as st1_...

Each .story or .scenario file must not exceed 20 MB. 1667 also applies these structure limits:

  • A MessagePack array or map in a current .story file must not exceed 50,000 items.
  • The fragments list in a legacy JSON .story file must not exceed 50,000 items. The command refuses a file that exceeds this limit.
  • If the legacy datablocks list exceeds 50,000 items, the selected story line imports without its retry history. The Fidelity Report lists the retry history as malformed.
  • Decoded MessagePack and JSON input must not exceed 500,000 values.

The selected story line must fit within 5,000 parts and 4,000,000 UTF-16 code units. Retry takes use the remaining capacity. If they reach either limit, the selected story line imports without the remaining retry takes. The Fidelity Report lists this omission.

The last value is the story ID. Use it to open the story:

1667 --story st1_...

A .story file is a Container. It carries the prose, the embedded Lorebook, the Memory, and the Author’s Note. The Memory becomes one always Fact with the memory tag. 1667 tries to turn each enabled Lorebook Entry into one Fact. Empty or invalid text and the story’s Fact limit can prevent this transfer. Check the Fidelity Report for skipped or changed entries. The Author’s Note text stays the Author’s Note.

A .scenario file carries a prompt in place of prose. The prompt becomes the first story parts. The Memory, the Author’s Note, and the Lorebook import the same way.

Import reads Scenario versions 0, 1, and 3. It reads NovelAI Lorebook versions 1, 3, 4, and 6. The command refuses an unknown Scenario version or a standalone Lorebook with an unknown version. If a story or scenario contains an unknown Lorebook version, the prose imports without its Lorebook Facts. The Fidelity Report names the omitted Lorebook version.

You can give more than one file. If the command cannot read one file, it continues with the other files. It prints each failure and exits with an error status.

Import a Lorebook

1667 import-lorebook adds Facts to a story that already exists. It does not make a new story, so --story is necessary:

1667 import-lorebook --story "Alderaan" world.lorebook

The command prints one line for each file:

world.lorebook: imported 12 facts into "Alderaan"

The command reads a .lorebook file as JSON or inside a PNG. It reads the file content to find the format. It does not use the file name.

The Lorebook JSON must not exceed 1 MB.

The PNG reader accepts uncompressed tEXt metadata with the naidata key. If a PNG has no archive metadata, the import reports no lorebook data in this PNG · export the lorebook again from NovelAI. It does not inspect image pixels or another hidden encoding.

To add a .lorebook or World Info JSON file to the open story, select import archive in the command palette. This command does not read a PNG Lorebook. Use 1667 import-lorebook for that file.

The in-app command can import .story and .scenario files, but it does not write their Fidelity Report. Use 1667 import for these files so that you can check the report.

The Fidelity Report

An import can change or omit data. The Fidelity Report lists the changes and omissions described below. It does not list Unicode normalization, line-ending changes in prose, paragraph splitting or trimming, title whitespace trimming, or title truncation after 4,096 Unicode characters. The import commands print the report to standard error, one line for each file.

Each .story import reports its retry takes and its one fixed omission:

alderaan.story: 3 retries imported as unselected takes; generation settings omitted

Each .scenario import reports the part count, the literal ${…} placeholders, and its own omissions:

mira.scenario: 3 prose parts; ${…} placeholders kept literally; description, placeholder metadata, tags, context defaults, bias groups, and generation settings omitted

Each Lorebook import reports the entry count, the Fact count, and the omitted NovelAI conditions:

world.lorebook: 14 entries read; 12 facts imported; unsupported search ranges, bias groups, and advanced conditions omitted

1667 does not read sampling settings from a NovelAI .story or .scenario file. Import a NovelAI .preset file with 1667 profile import. Refer to Generation Profile transfer. A .preset file must not exceed 64 KiB (65,536 bytes).

The report adds one item, after a semicolon, for each other change that occurred:

Item Meaning
memory truncated to 100,000 characters A Fact holds a maximum of 100,000 characters
2 disabled entries skipped A disabled Lorebook Entry does not import
3 entries truncated to 100,000 characters The entry text was longer than one Fact holds
1 entry could not be read The entry was not a readable record

Where your data lives

1667 stores your stories on your machine, in a .1667/ directory inside a project root that you select. Refer to Story storage.

  • 1667 has no account system and no cloud storage.
  • 1667 does not collect usage data.
  • 1667 stores provider secrets only in the machine tier. During generation, it sends the selected secret only to the configured provider.
  • 1667 sends your prose only to the configured provider. It sends the prose in the generation requests that you start, and in the token-count requests that keep the context meter exact.

You can export your 1667 stories as NovelAI Archives. Exit 1667 first. These commands export the most recently updated story:

1667 export --format story
1667 export --format scenario
1667 export --format lorebook

Use --story st1_... to select a story by its ID. Use --all to export every story.

These Archives are transfer files, not complete backups:

  • Of the available prose branches, a .story or .scenario Archive contains only the selected story line. Both formats also contain Facts, Memory, and the Author’s Note. A .story Archive has no retry history.
  • A .lorebook Archive contains Facts but no prose.
  • The exports can omit alternate takes, directions, summaries, chapter breaks, the first chapter title, tags, and Side Notes. The Fidelity Report counts chapter breaks but does not name a lost first chapter title.
  • The exports omit Fact secondary keys, secondary-key mode, scan depth, recursion, priority, per-Fact token budgets, creation times, and source story part links. The Fidelity Report does not list these losses.
  • The exports omit the Facts budget, Author’s Note depth, story phrase bias, and banned strings. Both .story and .scenario omit the Author Brief. Only the .scenario Fidelity Report names that loss. The exports also omit Image Attachments, thoughts, token probabilities, and Generation Records.
  • For each prose part, the exports omit the model name, whether you wrote the part, human-edit ranges, rewritten ranges, and creation and update times. The exports also omit the story origin.

The export command writes a Fidelity Report to standard error. Check it for changes and omissions. Refer to Story storage for the content and selection rules of each format.

NovelAI model access

1667 does not send requests to NovelAI’s hosted models. The project does not plan this integration.

Use an OpenAI-compatible host, Anthropic, or a local model server. 1667 supports chat and text completion protocols. The file import and export features do not depend on NovelAI model access.