What's new in Novelcrafter

Updated System prompts

By popular demand the system prompt for prose generation has been tweaked based on your feedback:

1. If the previous scene matches the same POV, it will now fall back to the last words of the previous scene to give the AI a head start for prose
2. Less words by default: The previous 900 words was too much and tended to cause the writing to drift too much. This has now been reduced to just 400 words (Again, these are just guidelines for the model. Actual word counts might differ).
3. Toppy has been switched to use the OpenRouter free model instead. This makes it easier on newcomers to try out a few functions without having to pay.

Model Updates

New Claude 3 Model Support

Claude 3 Opus and Sonnet have been just released and are now available inside of Novelcrafter as well (via OpenRouter).

Unmoderated Claude endpoints

All Claude models have now also been switched over to their unmoderated endpoints. This should result in less strict responses.

While Opus is a good model for writing, it is also very expensive! Please make sure you have the funds to use it.

Mistral Medium + Large have been added to system prompts

Mistral recently released their “Large” model and you can now use both, medium and large for various actions (chat, summarization, …) as well.

New /section Command

Instead of having to type text, select it and then put it in a section, you can now easily create a new section by using the /section command in your manuscript.

Other changes

• Fixed issues with commas in your chapters when creating an outline from text
• Highlights in the timeline will now be hidden for collapsed sections as well
• Various other bugfixes and improvements

Improvements

• Prose sections can now be colorized! 🌈

• Act/chapter/scene texts/headings in matrix view as well as the selected scene in chats are now links to the manuscript

• Add codex.getAliases("Entry name") to get a comma-separated list of aliases for the given entry.

• Add context.codex.getAllByType and codex.getAllByType to get all entries for character, location, object, lore, subplot and other

• Make code blocks in chat messages more readable

• Improvements to chat message extraction

• When calling moderation, you now see a little message about its status

Fixes

• Chat Codex Entry Tracker shows entries again
• Codex additions in sections are now also being detected
• Hide word counts in chat message when AI is idling
• You can now use {codex.get("Entry name")} on entries that have the “do not track” flag enabled as well

A small one today, mostly just internal bugfixes and improvements. In addition:

• Codex Entries with quotes in their name should now be detected properly in prose as well
• You can now “add from archive” in outline plan mode, too

Redesign of the AI Connection screen

Since the current settings screen was a bit confusing for newcomers (click here, click there), the new screen has been reorganized for less clicking around. This is only step 1 in upcoming `usability improvements in that regard!

Collapsible Sections in your Prose

Sometimes you generate multiple versions of a beat, sometimes you want to label things - or you simply want to hide things from the AI. You can now add sections within your manuscript to do just that!

You can create new sections from the bubble menu when selecting text, using the new “Section” button. Sections can be labeled, collapsed and also toggled on or off for AI (as in, should your prompts include the text).

Other changes

• New context.isStartOfScene function
• Added optional lookback limit to context.storySoFar function: You can now specify how many scenes it should combine (e.g. context.storySoFar(10) will return the previous 10 summaries)
• Added Filipino and Tagalog to novel language dropdown
• Fixed Ollama custom base URLs not being recognized
• Renamed “delete excess scenes” to “delete empty scenes” action to avoid confusion
• Other various fixes and improvements

The payment processor Novelcrafter uses just added support for Subscriptions via PayPal. That’s right; You can now subscribe without a credit card as well! There is one caveat to this that only applies to PayPal subs: When picking a plan you are currently unable to upgrade/downgrade, you can only change the billing cycle (monthly or annually). We already contacted the payment provider to provide further clarification on the details.

• Improve chat message extraction to find even more content
• Duplicate chat messages are now hidden
• Fix being unable to hide help bubble
• Fix word counts not updating in manuscript/write view

Some custom prompt updates as well:

• Add isLimited and isOmniscient checks to POVs
• Add string.contains, string.startsWith and string.endsWith to functions
• Add not to logic functions

And the usual various other bug fixes and improvements, too.

Chat Message Extraction

You can now easily extract content from your chat messages:

1. Scene Beats: When beats are detected, you can now copy them all into the clipboard to paste them into your scene - all at once!

2. Codex Entries: No more manual copy paste of your cast or settings. Let the system extract the details, and you can tweak them later.

3. Chapters/Outline: When the system detects a list of chapters, you can now add those to either a new, or an existing act.

To trigger the extraction, use the new “Extract” button below the AI response in the chat.

This feature does not use any AI calls and depends on internal logic. While many different cases have been implemented, there can be misses. If the system doesn’t detect anything, please ask the AI to rewrite its output in a list format.

New Custom Prompt Functions

• New logic functions: and, or, isEqual and isNotEqual
• New checks against POV types (scene.pov.isFirstPerson, scene.pov.isSecondPerson, scene.pov.isThirdPerson)
• New context.actSoFar (since we already got storySoFar and chapterSoFar)
• The context.previousScene now has the same list of calls as the regular scene (e.g. context.previousScene.pov.type)

And as always: various other bug fixes and improvements

• Add ability to archive and restore custom prompts to clean up your list (but don’t get rid of them juuuuust yet)

• You can now specify which codex entry types a custom codex category should collect

• The character associated with the current POV will now always be added to the context (should help with 1st person scenes)

• Duplicate values are now removed when saving tags or aliases

• Make help bubble toggleable in user settings

Custom Codex Categories/Groups are here!

It is now possible to create dynamic codex groups based on a list of tags:

• Create a group from the gear-menu in the codex sidebar
• Each group has a scope (either the novel, or a series) and a list of tags that fall under that group
• You can also now easily assign a color to all entries associated with that group
• The groups show up in context menus and the plan matrix too!

New Plan mode settings have arrived!

They are now stored in the browser, just like the manuscript view settings, so you no longer need to set up your view every single time you open a novel!

• Toggle auto-detected references on or off!
• Change scene card height between full (height determined by summary length), and other fixed sizes
• Matrix view now also respects those settings (e.g. card width and height)

In addition

• New Subplot codex category. More to follow for this one!
• Your labels are now sortable by hand, instead of alphabetically.

Very small one, but still a couple improvements:

• When using the chat, the message box should now regain focus once the AI did its thing. No more clicking around!
• Scene subtitles are now visible in matrix view
• New {context.codex.global} function to only list entries marked as “always include”
• A few other bug fixes and improvements