The AI Settings tab contains a number of important and cosmetic options: Text streaming, sentence trimming, and HypeBot, a commentating companion designed to offer fun input on your story! Experimental features like token probability display and highlighting, context length, and more are available in the Experimental section.
The AI responses section contains token streaming and sentence trimming options. These settings are kept enabled by default for increased readability.
Stream AI Responses
Enabling Stream AI Responses causes the text to appear on screen token by token when generated, rather than having it appear all at once. The Streamed Response Delay slider below adjusts how quickly the text is streamed, with lower values being faster speeds.
Continue Response to End of Sentence
With this toggle disabled, the AI continues to generate until your set Output Length is reached, which results in outputs ending mid-sentence. Enabling this causes the output to end on periods or other sentence-ending punctuation.
HypeBot was created for the 1st anniversary of NovelAI, It uses a unique model with data provided by our Community Manager Aini and a flurry of our amazing writers. Hypebot has seven selectable avatars based on our model characters and our mascot: Goose! You can even import your own using the Custom CSS feature in the Theme tab!
The HypeBot appears at the bottom of your screen during a story and offers witty, unapologetic encouragement—and sometimes things more unexpected...
The Automatic Setting causes HypeBot to only appear when triggered by the Comment Chance percentage, and disappear when another output is generated. Setting Comment Output to Permanent allows the HypeBot to remain on screen at all times.
A number of avatars based on the NovelAI family of characters are available for choosing under the dropdown, but you can also set your own using an advanced option in the Theme tab.
Goose Tip: Don't worry, the selected avatar has no effect on HypeBot outputs, they all use the same AI model! Pick whoever you want!
This slider determines how often HypeBot pops up to speak, if Comment Output is set to permanent, the previously generated response from HypeBot will disappear. Setting this Slider to 0 is the same as disabling HypeBot, while setting it to 100 percent will make it generate a response with each output.
Comment Streamed Response Delay
Much like the Delay slider in the AI Responses section, this slider determines how fast HypeBot text is streamed.
Goose Tip: Clicking on the Hypebot avatar in the Editor generates a new Hypebot output!
The Experimental section contains settings designed to effect AI output and direct the AI towards story generation, as well as enabling Token Probabilities, which allow you deeper insight into your generations! The features in this section might be unfinished or changed drastically in the future, but don't let that stop you from checking them out!
The Preamble toggle, as described, adds an asterism (⁂) to the top of your context, a formatting method that tells the AI it is the start of a story.
Goose Tip: Consider keeping this option disabled when using the AI to generate non-story formatted text like chats, lists, or stat breakdowns.
The Trim Excess Whitespace toggle removes trailing spaces and repeated newlines that might cause issues with generation before sending your context to the AI.
Depending on your subscription tier (1024 for Tablet, 2048 for Scroll and Opus) the Force 1024 token limit toggle allows you to switch between a context size of 1024 or 2048 tokens. A smaller context might be beneficial if you want the AI to focus more on recent context and actions, or if you are writing prompts and stories for lower subscription tiers.
With the Default Bias toggle enabled, a bias of -0.12 is applied to reduce the likelihood of dinkus (***) and asterism (⁂) characters appearing, resulting in fewer story breaks and a longer story flow.
Goose Tip: While the Asterism (⁂) is typically used for story starts and titles, the dinkus (***) is used to separate a chapter or represent the passage of time!
Checking the Enable Token Probabilities toggle causes the icon to appear in the Editor at the bottom of the screen near the Retry button (or the Lorebook icon if using Editor V2) after generating an output.
Clicking it opens the Token Probabilities window, where you can view an interactive breakdown of the most recent output. Selecting a token or token ID in this window allows you to view the percent chance that each token had of being selected.
Goose Tip: The colors used for Low, Medium, and High Probability are visible and customizeable in the Theme tab!
When using Editor V2, or the inline generation hotkeys (Shift + Ctrl + Enter/ Shift + Command + Enter) the Bidirectional Inline Generation toggle switches between using a special model that takes both sides of the context into account when generating, or the model you have selected in the Defaults tab.
If Editor V2 is enabled, this checkbox highlights the most recent outputs token probabilities in the Editor window. The colors used are the same as in the Context Viewer and Logical Probability Viewer, color coded based on their probability using the colors set in the Theme tab.