Changelog
Follow up on the latest improvements andΒ updates.
RSS
improved
fixed
Plan
Write
Review
Prompts
September 5, 2024
Word counting changes
This update makes word counting to be more in line with Scrivener. Both, Word and Google Docs (which we have used as a guideline so far) have issues counting words with em-dashes, punctuation and other special characters. The updated word counting is now more like what people would count as real words.
For comparison:
- I wish you would - oh, never mind.now counts as 7 words, not 8.
- It was β¦ complicated.now counts as 3 words, not 4.
Performance improvements
We internally upgraded to a new version of the framework Novelcrafter is using to render stuff on screen. While there were great performance improvements, the memory consumption should be even less for bigger novels now. Though we did extensive testing, as always, please report if some things aren't updating quite right.
Update to Mistral models in system prompts
For all system prompts, we now configured all Mistral models to go through the "Mistral" provider on OpenRouter. While OpenRouter does allow calls to "Azure" as well, those have a special moderation layer that can cut off the output too early if it detects an issue, causing you to loose money as the AI already started responding. Novelcrafter cannot show you when these "in-stream moderation" errors happen yet, so we opted for this solution for the time being.
Other Changes
- Added the ability to copy all scene beats into the clipboard (for scenes, chapters and acts)
- Added a "Hide my email" setting for YouTubers/presenters/teachers that don't want to have their personal mail on display.
- Fixed some codex detection issues on older versions of Safari. We're now using a fallback that should work OK for the most part. If issues still arise, please update your browser.
- Fixed not being able to dissolve collapsible sections
- Fixed chapter titles (and automatic numbering) not always refreshing/updating and being in an inconsistent state
- Fixed a few paragraph indentation inconsistencies with beats/additions and sections
- Fixed some visual shifting when showing a horizontal cursor before a beat at the start of a scene
- Fixed codex thumbnail uploads saving to the wrong entry on slow connections, when immediately switching to another one
- Fixed context.previousBeatalways picking first beat of a scene, instead of the one before
improved
fixed
Write
Review
Import
Collaboration
August 31, 2024
This release mostly contains bug fixes and general improvements:
Manuscript improvements
- Added some mitigations for the wild cursor jumping. While not totally gone, we hope these changes help in some cases. Please keep the reports coming if it's still occurring!
- The scene beat validation will no longer nag about entries with nested references having no description, or that your global entries have short descriptions. Since these are common use cases, it makes no sense to have a constant warning for those.
- Added a selection outline to scene beats and collapsible sections to indicate if they're focused (e.g. when you used the drag handle to move them around)
- Fixed not being able to put a cursor between a scene beat and a following section
Other changes
- Added an additional "Add Chapter" button to the end of Acts when in the Outline plan view
- Added the ability to switch the codex "filter by tag" to "match any", instead of requiring entries to match all selected tags
- Fixed some errors with .docximports where formatted text was incorrectly interpreted for headings and other parts
- Fixed some issues around handling Asian languages (like Thai)
- Fixed some visual inconsistencies with sharing novels and series depending on your plan and/or team status
- Fixed the "characters per scene" chart sometimes not loading correctly
- Some optimizations around app loading times and general memory usage (e.g. previously, prompt previews were kept in RAM indefinitely until reloading Novelcrafter)
- Fixed the "you're no longer logged in" modal appearing on App start when on a slower connection
- Added a max-width to menus for when names get too long (e.g. series selection)
- Some additional improvements for chat extraction with heading + blockquote cases
- Various other bug fixes and improvements
improved
fixed
Codex
August 23, 2024
This release changes the routing to Novelcrafter itself, which should increase stability and reduce latency across the globe, especially for people in Oceania or in rural places.
In addition, the following minor changes have been made:
- Fixed some issues around codex matching for entries with quotes in their name
new
improved
fixed
Codex
AI
Teams
Collaboration
August 19th, 2024
Teams for your friends, family, colleagues and more!
To create a team, you need the Specialist plan, but everyone else can benefit as long as they have any plan!
![image](https://canny.io/images/78aeb143ca0f8e372407d4e6e2ea5835.png)
Let's say you're working on more than one book or series, and you have to add a new person to your writing projects. Before now, it was quite tedious to add them individually, but with the new teams feature you can set one up once and simply share your novels or series with everyone involved!
You can reach the
Teams
feature through the settings dialog.You can invite members by email, and even have everyone set whether they want to be visible to the others. Team leads can also decide if sharing of novels and/or series with the team should be allowed and even hide the member list entirely (e.g. for more "public" teams with dozens of members).
With the introduction of teams we also reworked how invitations work and look; only owners and team leads can now see the emails of invited people/members. For privacy reasons we've disabled showing them for anybody else.
Shared with me - At a glance
![image](https://canny.io/images/bdd236a9bd17b3bf242d62ed6c120e57.png)
For those that have a lot of novels or series shared with them, there was no easy way to figure out who owns what or where you're all involved with. The new "Shared with me" tab in the settings lists out all the content either shared with you directly, or through a team, and allows an easy way to leave them, too.
Move codex entries to... somewhere else?
![image](https://canny.io/images/479d583cc3f4bba1065152a82353c9b7.png)
Sometimes an entry doesn't fit into your current novel or series, or has been mistakenly added to the wrong project. The new "Move to..." feature allows you to move the codex entry somewhere else altogether!
Performance optimizations for large codexes
Some people have started to build up hundreds of codex entries and the performance for matching your entries has tanked quite a bit.
A new system has been built that should work much better with entries that share a common structure (e.g. "Town name - entrance", "Town name - central", or things like "Country - Flora & Fauna", "Country - Towns").
In addition, any archived scenes you have will no longer be looked through either, since it didn't make much sense in the first place.
Other changes
- Codex additions now have a double-newline between them when sent to the AI (previously, they were all concatenated together)
- Any dialogs left open will no longer be saved as part of the "remember where I left-off" memory. That was a bit confusing.
- Fixed deleting prompts leaving a local copy behind until page reload
- Some additional fixes to chat/snippet extraction for weirdly formatted lists
- You can now open the series page from the codex series header
- Added a little list of providers for the OpenRouter provider routing settings in custom models
- Various dialogs (like settings or revision history) on mobile now have a fixed height and some other improvements with regards to usability
- Various other bug fixes and improvements
improved
fixed
Chat
Codex
Prompts
Snippets
August 13th, 2024
Archive (almost) everything
Novels, codex entries, snippets and chats can now be archived and restored. Don't want to delete them right away, or declutter your content? You can now archive them and bring them back any time.
If you want to see your archived content, look for the funnel/filter button:
![image](https://canny.io/images/decec3600661fb7ca8df8f8f65eec385.png)
You can see a detailed guide over on our documentation: https://docs.novelcrafter.com/en/articles/9733992-archiving
This also means that archived novels are hidden by default on your home screen (clear the filter, or toggle to "archived" to view them again).
Custom prompts
After some further testing, we are now including the name of the character for each scene in the novel outline. When using chat you can now talk about his & her stories or other setups with either single POVs or alternating viewpoints!
This release also fixes indentation issues in prompts - any indentations or extra spaces at the start of the line are now kept all properly (this should improve AI prompt adherence in some cases). This is helpful when you have multi-level markdown lists for writing rules in your prompts.
Other changes
This release also contains a couple important bug fixes:
- Fixed the 'unknown' environment banner appearing when it shouldn't
- Fixed the codex matching not working after weird whitespaces, or at start of lines within chat
- Fixed "continue where left off" novel covers being too big on the home screen for mobile users
- Fixed custom codex categories still being visible after deletion (until reload)
- Some other little bug fixes and tweaks
improved
fixed
Chat
AI
Import
August 5th 2024
Novel Outline Updates for Chat
For those who have experienced issues with AI not recognizing or miscounting scenes/chapters, this update should significantly improve the experience:
The internal structure for the "Include novel outline" feature has been updated to use a more structured approach (XML instead of Markdown). This should
help guide the AI when discussing specific acts, chapters, or scenes
. We've also made scenes with content but no summary visible to the AI, preventing it from underestimating the number of scenes in your story.A new use case now possible:
- Create a new, empty novel
- Apply a template from the "Create from Outline" feature
- Open a new chat and select "Novel Outline"
- Ask the AI to populate your outline with an idea
- The AI should now use the exact act and chapter names, as well as the precise scene count from your plan view
If you prefer the old behavior for your custom prompts (although we consider this a major improvement), you can change your
novel.outline
call(s) to novel.outline("hierarchical")
.Other chat updates:
- The Chat prompt preview no longer requires a messageand works with a blank message
- Changing the prompt and/or model for a new chat thread now properly saves it as a new chat (this was already the case, but empty chats were only visible after refreshing the page or reopening the novel)
- The chat will now show a little "knowledge cutoff" warning so you know which messages will actually be sent to the AI.
- Fixed some minor issues with chat extraction involving nested lists
- Resolved interactivity inconsistencies when accessing chats in shared novels or subscription plans without workshop chat
Import Improvements
We've updated the novel import dialog, including a new setting to enhance imports for users who only use scene dividers. New help text has been added to better explain the options.
![image](https://canny.io/images/ae8027e71fb78a682d5f4124d63d36d4.png)
For those that don't want to bother with titles and the like, and only have a single-chapter novel, the importer now supports those as well!
Other Changes
- Fixed the "[...] maximum context length [...] However, you requested about [...]" error caused by a recent internal change in OpenRouter
- Resolved an issue where an extracted chapter, when added to your outline, would not save correctly and would lose its association with the act upon page refresh
- The export button now always appears when the sidebar is collapsed
- Implemented additional performance optimizations for writing in scenes
- Fixed visual issues with some dropdown menus not overlaying content properly
- Various other fixes and improvements
improved
fixed
Write
Prompts
July 31st 2024
π»ππ πππππππ ππππ: August Writing Bingo π»π
We got something exciting to share with you: going forward, every month, we'll send out a new, interactive theme machine for writing prompts and more.
For August, we prepared a fancy Bingo grid. It's there to help you find new and exciting ideas and hone your writing skills along the way. Choose a row or column yourself, use the randomizer, or take a chance with the chaos button! Play around with it here: https://monthly-muse.com/1 (you can also reach it via the NC website.)
If the theme machine doesn't come up with something great, we also included some creative sparks for more fresh ideas.
Updated Model Lists
We cleaned up the current model lists for the system prompts and
finally added GPT-4o mini
! This should make longer chat sessions a lot more budget friendly.In addition, this release includes the following bugfixes and such:
- The "start writing" placeholder should no longer appear where you couldn't actually start typing
- Fixed an internal editor issue where any ellipses and em-dashes would not be replaced properly
- Fixed an issue with when entering simplified Chinese into text boxes, the text would appear as traditional Chinese.
- Fixed some layering glitches with the scene actions menu in the manuscript page
- Various other bug fixes
new
improved
fixed
Chat
Prompts
AI
July 29th 2024
Chat Prompt Preview
The long awaited chat prompt preview is here! You can now see what gets sent to the AI when entering a question/message. You can access the prompt preview button from the model switcher in the chat thread header (next to "Select AI").
![image](https://canny.io/images/6904990edf1337cdbc14db65ddcff7c3.png)
Custom Prompts
For some, the custom prompts can be a bit daunting to look at when unfamiliar with programming languages. So we added support for comments inside your custom prompts! The system default prompts now have a couple to tell you, what each section does.
To wrap text in a comment, wrap it in
{!
and !}
, like so: This is some text {! and this is a comment !}
.That way you also can just disable certain functionality from your prompts, without having to remove it altogether.
- Added support for underscores in numbers, so you can now write 2_000to make it easier on the eyes
- Added support for 'dangling' commas in prompt functions, so it no longer breaks when you write codex.get("name",)
Other Changes
- Fixed a bug where older codex entry versions might have been used for AI generation.In case you got weird output lately, this might have been the cause.
- Nested entries of codex entries directly attached to a scene will now be included as well. This is how it always should've worked, so we count this as a 'bug fix'.
- We now cache the OpenRouter model list so that, in the case there is downtime/are issues, we should have a working copy (fingers crossed!)
- The default scene summarization prompt has been tweaked to not glitch into the scene's POV for writing (it should now always write in third person, present tense.)
- Other various fixes and improvements
improved
fixed
Write
Codex
AI
Export
July 27th 2024
Minor updates to the novel exporter
![image](https://canny.io/images/60d29e36cbc4a65c129885a0e3ad7eb1.png)
- Scene Dividers can now be set to "none", "asterisks" and "headings". The latter is new, and simply adds a new heading with the scene number
- For those with a lot of scenes/chapters, the export window is now scrollable, so it's easier to quickly export your book in one go
Other changes
- The "continue where you left off" function will now actually remember your last state. No more re-creating your previous set up!
- Any search box now holds for a bit until you're done typing. This should improve performance for those that have a lot to crawl through
- Fixed some issues around sectioning off AI generated text
- The editor will now reserve an empty paragraph after sections, beats and more. Sometimes you couldn't get out of one of these boxes.
- Fixed a special case where the "+"-button next to a codex heading would not open a new panel (for novels that aren't part of a series)
- Other various fixes and improvements
new
improved
Write
AI
July 22th 2024
Performance updates
This update contains a
ton of performance updates
that reduce the number of updates the app does as you work. Editing codex entries should now be smoother/faster, as well as using AI actions should now be a lot faster as well (less choppy/lagging).Manuscript scene hopping
While you were able to say "remember where I left off", that only applied to the exact scene and position, but not how your write view looked like. So going back to it, you always ended up back with "everything". This was annoying. So we fixed it.
You can now always go back to exactly where you left your manuscript, which should help a lot with larger projects.
AI interactions revamp
![msedge_nr3qnsPlRP](https://canny.io/images/c90d608e078afa229171e26de7a8eb5c.gif)
As some of the performance issues stemmed from the AI interactions, this aspect has been updated and visually overhauled, which in turn made a couple goodies possible!
- You can now undo/redo any AI text replacements! This also means that if you hit "retry", you can now go back to a previous generation in case you liked that one better.
- You can now see your generated word count. This will only show the generated words, so if you you start to cut and edit the text down, this won't update.
- There's a new button to stop your generation. You could already do this by hitting "Apply" pre-maturely, but that also... applied the text. So this new "Stop" button gives you a bit more choice and clarity.
- Another new button to put your generated text into a section. By default, this uses the model name that generated the text as its title.
And yes, this update also fixes the issue of the apply/retry/discard buttons being stuck or keep floating around. That was indeed quite annoying.
Other Updates
- When opening just a specific scene in the manuscript, the manuscript switch menu will now show which chapter you're currently in (the relevant chapter will have a little dot next to it)
- The listing for models on OpenRouter will now retry intelligently, if there was an issue grabbing the data
- Fixed a couple issues when a prompt function does not receive a required argument
- Other various fixes and improvements
Load More
β