Wow! It has already been a whole year since NovelAI was released!
So, we would like to thank you all for your support.
The team thanks everyone for their continued support and love throughout this whirlwind year.
It started with the first spark of the team coming together, to the Discord exploding over the course of a few days, through the ups and downs.
Please accept our thanks for every time you have peeked into the Discord, participated in the chat, shared your NovelAI content and memes or stopped by to check out the social media accounts. You’ve shared your feedback, wishes and most of all your presence and, you’ve shared the word about NovelAI. You’ve blessed us with your questions, comments, requests and feedback!
It’s all got us where we are today: the next wave of features and updates coming to NovelAI!
Some of the things announced during the stream have already arrived in their earliest iterations, so let’s cover them first!
HypeBot — Released
Get hype with HypeBot! First presented during the 24 hour Twitch stream we’ve created a very enthusiastic companion to support and encourage you: HypeBot! This companion will comment (with relentless enthusiasm) on your stories and give you the occasional suggestion to help move your plot forward.
You don’t have to face your stories alone any longer!
Count on our faithful companion to provide you with enthusiastic commentary, suggestions and support through thick and thin (no matter what you write about). Adjustable appearance settings, TTS compatibility and smooth animations and transitions make the Hypebot a fun addition to the editor.
Hypebot is the first step in the direction of custom trainable bots to accompany you or others through your adventures in the future. There are a lot of ideas the community raised during the 24 hour twitch stream and since the release and we’re excited to further improve upon this novel little feature!
Also, If you don’t like it you can easily disable all Hypebot features in the User Settings.
We added custom CSS tags for you to easily customize the images of the Hypebot! You can add life to your own Hypebot by editing the idle and loading states of the Hypebot portrait image using theme CSS tags.
The Custom CSS Tags you can use in the Theme Settings Custom CSS dropdown to customize the picture of your Hypebot:
comment-avatar Always present on element. Can be used to set an image that will be used regardless of state
comment-avatar-speaking The element has this class when the text box is visible and not doing the “…” thing
comment-avatar-speaking-active The element has this class when streaming is in progress, and for a variable (based on generated characters) amount of time after a generation completes
comment-avatar-thinking The element has this while generating with streaming off, or before the first token when streaming is on. This is where it does the “…”
comment-avatar-idle The element has this class when there is no text box. This only happens when you have it set to permanently show up.
comment-name Can be used to set the name via the ::after pseudo class e.g .comment-name::after { content: ‘Name’ !important; }
Hypebot CSS Example
If you wish you may copy this into the Custom CSS field on your theme and simply replace the Name you wish to display and the image file links that you want to use. Keep in mind, images need to be urls and can not be local files.
.comment-name::after { content: 'NAME HERE' !important;}.comment-avatar { transition: background-image 0.1s ease-in-out; background-image: url(https://EXAMPLEURL/IMAGE.png) !important;}.comment-avatar-speaking { background-image: url(https://EXAMPLEURL/IMAGE.png) !important;}.comment-avatar-speaking-active { background-image: url(https://EXAMPLEURL/IMAGE.png) !important;}.comment-avatar-thinking { background-image: url(https://EXAMPLEURL/IMAGE.png) !important;}.comment-avatar-idle { background-image: url(https://EXAMPLEURL/IMAGE.png) !important;}
Some details on the training of the Hypebot as shared on the stream: Aini trained this hypebot with a total of 396 (very enthusiastic) comments, made on a large variety of different story context snippets. The snippets varied in perspective, tense and content and this process took about one day to create them. Comments were placed directly into an interface built into an internal version of NovelAI.
The Hypebot currently implemented into NovelAI is only one model, there is no difference in behavior if you select another AI Model display picture. We’ve had some reports of people thinking they are different but that is 100% placebo (at least until we implement more bots further down the line).
At this time we don’t have any estimates for Custom Hypebot training but we’ll make sure to keep everyone updated on any developments!
Also, as mentioned before, there have been some very interesting ideas for future Hypebot implementations, such as two type of bots commenting on the same story at once etc.
But we’re always excited to hear your ideas and wishes, so feel free to drop by on our socials or on our official Discord to share anything you might like to see in the future!
Inline Generation — Released
Our Inline Generation now actually considers the content that comes before and after the spot you select to generate. It will attempt to connect the dots between your text. However, it does not take into context the text you select in order to generate (if you do select any in the first place).
The original Inline Generation only let you select a new position for the AI to generate, anything past that point was never taken into consideration.
At this point we are aware that the name might cause some confusion and we are actually considering renaming the feature to something along the lines of ‘Rewrite’.
Text to Speech V2 — Released
A new era of NovelAI’s streamed TTS has arrived! With TTS V2 we added a large number of much more unique and expressive voices to the TTS system to extend its capabilities. The potential for finding new unique voices has been vastly enhanced! — It is now also possible to control TTS V2 seeds on a whole new level with seedmixes.
Seedmixes allow you to mix and customize voices in plenty of ways — for example, try combining two of our defaults with seedmix:Aini+Pega, or set specifics with seedmix:Aini|intonation:Lim-Naia|cadence:Raid|style:Orea.
Prefix seedmix:
Anything after you can do basic math with seedsseedmix:seed1+seed2+seed2 etc.
There is no limit to the amount of additions or subtractions, but it is important to be careful not to subtract too much, otherwise the outputs might turn undesirable, also keep in mind there is a 256 character limit.
You can specify different seeds (or combinations of seeds) for different parts of the TTS.
Cadence: the flow of words as the voice speaks, the pauses and breaks and breaths.
Intonation: the peaks of the voice and lows, how animated the voice is
Style: the tone/cleanness of the voice
An example of a seedmix that uses these:seedmix:cadence:Amy22|intonation:amy17+amy14-bob4+amy25|style:amy17+amy14-bob4+amy17+amy25This creates the voice amy17+amy14-bob4+amy17+amy25 but with the cadence of Amy22, which has a much more natural way of speaking.
Basic special syntax example: seedmix:amy-bob+alice to add/subtract voice characteristics.
Advanced special syntax can separately specify cadence, intonation and style: seedmix:Aini|cadence:bob55+bob or seedmix:cadence:alice+amy55|intonation:bob|style:eve or any other combination.
Limitations:
- Spaces don’t work with seedmix seeds
- Max length of seeds is 256 characters
Feel free to share your findings with the community!
We’ll be keeping a close eye on your creations and may even add them to the Default roster!
You might have also noticed a familiar voice… seed: Aini!
Our community manager sat down and recorded quite a lot of voice clips with corresponding text (1085 at this point) to let us experiment with our TTS training. After many iterations we’ve finally settled on this one which we think resembles her original voice quite closely.
Okay enough about already released content!
Let’s cover some of the coming soon:tm: stuff!
Editor Overhaul & Text formatting
This has been a long time coming! The Editor infrastructure is being completely revamped in order to future-proof NovelAI and resolve issues such as performance lag with long stories.
The new Editor will come with the following new features and changes:
- Text formatting! You will be able to format text as bold, italic, underlined, strikethrough and potentially more.
- Dynamic loading of story sections. With long stories, older sections of the story will be unloaded from the editor until scrolled up far enough. Similarly, only the relevant sections are loaded when opening a story for the first time. This should drastically improve editing performance!
- A editor toolbox for quick access to text formatting and other options. It will appear above selected text and allows to change formatting, inline-generate and more, both on mobile and desktop devices.
- Live detection of text origin. User- and edit-colors will appear and update instantly while typing. Edit-detection should also behave more sensible based on context.
- Undo/Redo in the editor is now integrated with the story Undo/Redo tree. For example, CTRL+Z will also undo the last edit or the last generation when appropriate, CTRL+Y will act like the Redo button, and such.
- Saving happens without having to leave the editor, and instead after a short pause on every change. Saving can also be manually triggered with CTRL+S. This should prevent loss of data when closing or reloading the browser-tab after a long editing session.
With the new editor being based on a new and improved Story data-structure, and both formats currently being incompatible, existing stories created with the current editor will continue using the current editor instead of Editor V2. We will investigate a conversion option in the near future.
Editor V2 will become available as as experimental option in the next couple of days, and become the default after making sure that it’s stable, battle-tested and feature-complete.
Image Generation
As AI powered Image Generation is currently en vogue, we realize the time is right to introduce our own self-hosted NovelAI Image Generation: without long generation queues curbing creativity, full encryption and as always without content guidelines or filters and in true NovelAI fashion, with a focus on generation speed.
At this point we plan to release our self-hosted Image Generation outside of the text editor first, in its own modal. In editor AI Image Generation that would display images alongside your story text is planned for release at a later point, when the editor overhaul has been rolled out completely.
The plan is to include available models such as Craiyon, a handful of diffusion models, as well as our own trained models.
Modules V2
Modules are getting a big overhaul! Our RnD team developed a new architecture for modules which is significantly improved over Modules V1. Now it can learn a lot more information, a lot closer to full model finetuning!
ModulesV2 is planned to be released this summer. With ModulesV2, Krake Custom AI Modules will release as well.
Additional Stream Topics:
Developer Mini-Interviews:
To help you get to know the faces behind the service we decided to create tiny interviews for each team member. It’s a little tough to get ahold of every member to fill their information out at this time but we managed to cover TabloidA, Chris (WIP which will be released in an updated version later) and Chims interviews live on stream!
Fingers crossed we can gather the last bits of information missing to host a full series of these or possibly finish going over the other team members in one of the upcoming live streams on Twitch! We’ll make sure to repost them all in their own blog post at some time as well.
Explore Profile Pictures
Explore Accounts is planned to feature profile images. So we tasked Aini to see if she could create a variety of possible characters for you to select as your profile image.
These are very early draft versions and are likely to transform quite a lot before inclusion if that even happens in the first place, so keep that in mind as you take a peek!
All of them are somewhat supposed to fit into the same mythological time and realm, but Aini is considering giving some of them alternative versions to represent some more animalistic sides to this roster. Initially the goal was to paint the 5 male and 5 female TTS nymph and dryad representations but considering the size of our user base we quickly realized that we would need a broader spectrum of choices. Don’t worry, we will include many more male options as well — the main focus for now was to quickly create a lot of options and to refine those later. Aini is also just a lot more confident in drawing girls, her approach so far is to get the easier task out of the way first.
We hope you enjoy the already released and upcoming features for NovelAI!
Please continue sending us your feedback, wishes, and suggestions throughout our second year!
Thank you for your support,
The NovelAI team.
