Add instructions here for the AI to follow. This is a good place to specify the writing style the AI should adhere to. For example, "Write in the style of Stephen King." or "Make sure to keep the tone of the story light-hearted and focus on dialogue instead of action."
The character limit for the author's note section is influenced by automatic memories made by the AI. The longer the story, the lower the limit. In the case of some models, you can increase the limit via the "Memory Boost" feature. Below you can see the current usage:
Author's Note and Memory Section usage:
Add information here for the AI to remember. If you have "Automatic Memory" toggled on, the AI should remember past events in the story. However, if you find it forgets some past story events, this is a good place to mention them. If you would like the AI to remember the world lore (descriptions of locations, characters, cultures etc.), then we recommend using the World Lore feature for this.
When adding things to the memory section, organizing them into bullet points or paragraphs may help the AI. For example:
- Emily discovers an ancient map hidden in her grandmother's attic.
- Emily embarks on a journey and meets a mysterious guide named Kael who offers his help.
- Emily and Kael find the hidden treasure, but they must confront an ancient guardian to claim it.
The character limit for the memory section section is influenced by automatic memories made by the AI. The longer the story, the lower the limit. In the case of some models, you can increase the limit via the "Memory Boost" feature. Below you can see the current usage:
Use this toggle to enable/disable the ability to generate NSFW content.
Note: If a model generates NSFW content, while the "NSFW" setting is toggled off, you will still be charged the full price for the generation, but the content will not be displayed.
Enabling adventure mode will make the experience more like a CYOA (Choose Your Own Adventure) story. It will make it less likely for the AI to generate actions that were not specified in the action input box. In other words, the AI will be less likely to "take control" of your character for you. The style of the text generated by the AI may also slightly change.
Important: If you're still having issues with the AI taking control of your character and performing actions you did not specify, then try enabling "Enhanced Output" and pasting the appropriate guidelines into the "Enhanced Output Guidelines" textbox, which you will find in the "Enhanced Output Guidelines" hint popup (press the "?" button).
Turning this on will make the AI automatically remember past events in the story. However, it may be useful to turn off "Automatic Memory" if you intend to place a lot of information in the memory section.
We generally recommend leaving "Automatic Memory" turned on, as it typically improves the quality of generated text. However, it may be useful to turn off "Automatic Memory" if you intend to place a lot of information in the memory section.
We generally recommend leaving the "Automatic Guidelines" setting on, as it typically improves the quality of generated text. However, it may be useful to turn off "Automatic Guidelines" when you are having trouble getting the model to follow instructions in "Author's Note".
We generally recommend leaving the "Automatic Guidelines" setting on, as it typically improves the quality of generated text. However, it may be useful to turn off "Automatic Guidelines" when you are having trouble to get the model to follow instructions in "Author's Note".
Turning on "Enhanced Output" can reduce repetitiveness and increase the overall quality of the text generated by the AI. This is achieved by having the AI rewrite the text each time.
You can also customize the "Enhanced Output" process to achieve various things, such as ensuring that the AI does not "take control" of the characters, whose actions you wish to control in the story. To do this, you can specify the guidelines the model should follow when rewriting the original text, as well as choose which model should be used for the rewriting process (check out "More Settings").
Please note that "Enhanced Output" can increase the time it takes to generate text, as well as increase the cost per generation. You can check the current cost per generation by clicking "?" next to "Model" in the settings bar.
Turning on "Enhanced Output" will increase the credit cost per generation, as well as the generation time.
Turning on "Capability Boost" can increase the quality of the generated text, by allowing the AI model to generate 8 story continuations simultaneously and then choose the best one.
Please note that the "Capabilty Boost" feature is available for select models only. Additionally, this feature can increase the time it takes to generate text, as well as increase the cost per generation by up to 400%. If a model is free, then using the "Capability Boost" feature will cost 10 credits. You can check the current cost per generation by clicking "?" next to "Model" in the settings bar.
Turning on "Capability Boost" will increase the credit cost per generation, as well as the generation time.
Select the model, which should be used for the rewriting process. Below you can see the additional credit cost associated with each model.
Magni 4.0 (Medium): 5 credits
DeepSeek V3: 8 credits
Aion-RP 1.0 (XS): 0 credits
Aion-RP 1.0 (S): 5 credits
Freya 4.1: 0 credits
Magni 4.5 (Small): 0 credits
Astral 1.0 (Small): 5 credits
Astral 1.0 (Medium): 10 credits
GPT-4o mini: 3 credits
Note: this setting is only available for stories that have "Enhanced Output" toggled on.
Provide custom guidelines for the model to follow when rewriting the original output. Please note that providing custom guidelines here will remove the default guidelines, which encourage the model to reduce repetitiveness.
Here are the default guidelines: "Replace any overly repeated words/phrases to reduce the overall repetitiveness."
Here are guidelines you can use to ensure that the AI does not "take control" of your character: "Make sure the character only performs the following action: "{{action}}". The text should be cut off at the moment this character begins to perform a different action, that was not specified."
Note I: the "{{action}}" is automatically replaced with the current action by the app.
Note II: this setting is only available for stories that have "Enhanced Output" toggled on.
Here you can choose the model that should perform the iterative refinement process. Each iteration of the refinement process consists of the following steps:
1. Analyze the strengths and weaknesses of the generated story continuation (either the initially generated story continuation or the story continuation generated in the previous refinement iteration).
2. Suggest improvements to the story continuation.
3. Generate a refined story continuation based on the suggested improvements.
Please note that changing the refinement model is only possible after "Refinement Iterations" has been set to 1 or more.
Current additional cost:
Here you can choose the number of iterations in the refinement process. Each iteration of the refinement process consists of the following steps:
Please note that setting "Refinement Iterations" to 1 or more will unlock the "Refinement Model" setting, as well as increase the generation time and cost.
Enabling the refinement process can significantly increase the cost per generation. You can always check the current cost per generation by clicking the "?" button next to "Models" in the settings bar.
Turning on "Memory Boost" will increase the memory of an AI model. This means the AI will remember further back into the story, however, this can come at a higher credit cost per generation.
Please note that this feature is available for select models only.
AI memory usage:
Turning on "Output Length Boost" will allow the AI model to output more text per generation. The additional cost increase of using this feature is calculated with the following formula: base_model_cost * (0.2 * (n - 1)), where n is the "Output Length Boost" setting value. You can always check the cost of the next generation by clicking on the "?" button next to the "Model" setting.
Note: setting "Output Length Boost" to values above 1 usually results in the model generating more text, however, this isn't guaranteed. Models can sometimes decide on an earlier stopping point on their own. Some models do this more often than others and the prompt type (which can be changed in the settings) can sometimes affect this.
Limit the maximum number of sentences the AI will generate in one go (type in a number between 1 and 10).
Set the number of times the AI will generate text when the "Send" button is pressed.
Note: credits will be used each time text is generated. For example, setting the number of generations to 10 means that each time you press "Send", it will cost 10 times the amount of credits. However, the credits will be substracted gradually, after each generation is completed. In case you want to cancel at any point, you can just refresh the page or close the tab.
Increasing the temperature can be thought of as making the model more random when generating text. This can boost the model's creativity, however, setting the temperature too high will negatively affect the model's ability to reason, and could even prevent it from generating coherent sentences. The default value for "Temperature" is 1.
Increasing this value reduces the likelihood of the AI model generating repeated sequences. This can be useful for generating more diverse text, but setting the repetition penalty too high may negatively affect the model's ability to generate coherent text. The default value for "Repetition Penalty" is 0.
Select the model you want to handle the World Engine logic. The model chosen here can affect the credit cost of a single generation. You can always check the cost of the next generation by clicking on the "?" button next to the "Model" setting.
Note: this setting is only available for stories that use the World Engine.
Use this setting to choose the lore retrieval system.
Off - This option turns off the lore retrieval system.
V1 - This is the first version of the lore retrieval system. It is the fastest and the cheapest, but it is also the least accurate.
V2 - This is the second version of the lore retrieval system. It is more accurate than the first version, but it is also slower and more expensive.
Use this setting to choose the the promp type for a model.
Auto - Use the default prompt type for a given model.
Standard - This is the standard prompt type that simply instructs the model to generate a story continuation.
Prefill - This prompt type can reduce repetition and prevent the model from prefixing its response with "Here is your story continuation:". To achieve this, this prompt type automatically prefixes the model's response with "Here is your story continuation:" to make the model "think" that it already generated this text and therefore decreasing the likelihood that it wil generate it again.
Prefill+ - This prompt type builds upon "Prefill" and can reduce repetitiveness even further. This prompt type automaticlly prefixes the model's response with "Here is your story continuation:" followed by the the last paragraph of the story. This makes the model much less likely to repeat previous story text, since it "thinks" it already did it.
Tags - This prompt type instructs the model to generate a story continuation and wrap it in "<continuaton>" tags. This can help the model understand that just the story continuation itself should be inside the tags (without previous story text or any other text).
Few-shot - This prompt type provides the model with examples of how to continue a story. None of the examples contain repeated story text, and therefore the model should be less likely to repeat previous story text when generating a continuation.
Note: When using "Completion Mode" or "Planning Mode", this prompt type setting is ignored.
The Deep Realms app uses different prompts for models, depending on whether the story text ends mid-sentence or at the end of a sentence.
For example, if the story ends with "Anne picked up the", the app will use a so-called "completion" prompt, to ensure that the model completes the sentence, instead of starting a new one. This prompt differs from the "standard" prompt used when the story ends with a full sentence, such as "Anne picked up the sword.".
Typically, the app chooses which prompt to use automatically based on the story text. However, enabling this setting, will force the app to always use the "completion" prompt.
Turning on "Planning Mode" can increase the quality of the generated text, by allowing the AI model to reason and plan before generating a story continuation. This differs from the "Enhanced Output" feature, which allows the AI to rewrite the text after it has already generated it once.
Please note that this feature can increase the time it takes to generate text, as well as increase the cost per generation by up to 30%. You can check the current cost per generation by clicking "?" next to "Model" in the settings bar.
1. Set as default - make the current story settings the default settings for all new stories within this world.
2. Create preset - save the current story settings as a preset, which can later be loaded easily.
3. Manage Presets - select and manage presets. On mobile, you can also access presets via the star button below the action input. On desktop, you can also load presets quickly by pressing the Alt key along with a number key (1–9).
4. View Changes - view the changes history of the story and revert the story back to its previous state.
5. Export - export the story settings into a file.
6. Import - import story settings from a file.
You can upload settings from a file containing JSON encoded settings. Use the Export button to see an example of the format.
Enter a name for the preset:
Are you sure you want to delete this preset?