[Full Local] I Built 'Narratable' to Weave Stories with AI Characters
The experience of advancing a story alongside multiple AI characters without connecting to the internet. To turn this ideal into reality, I am introducing the first installment of 'Narratable,' a tool I built myself as the operator of this account.
- Progress while protecting privacy in a completely local environment
- Draft a full set of world-building and plots from just one prompt line
- PWA support allows operation like an app even from smartphones
- Design that considers PC load, such as VRAM release upon closing
![[Full Local] I Built 'Narratable' to Weave Stories with AI Characters](/img/narratable-part1.png)
What is 'Narratable'?
'Narratable' is a roleplay chat tool where scenes featuring multiple AI characters progress with included narration. It is a unique system built by the operator of this account using AI (Claude) for personal use.
The biggest feature is that it is 'completely local.' All processing is completed within your own PC without using external cloud services. This allows you to construct a free world of stories while remaining unaffected by internet conditions and maintaining privacy.
- Simultaneous interaction with multiple characters possible
- Immersive direction including narration
- Executes both text generation and image generation locally
System Configuration and Operational Mechanism
Behind this system, several tools work in coordination. For text generation, 'Ollama (software for running AI models on a PC)' is used, employing the model 'gemma4:12b-it-q8_0'.
On the other hand, for image generation such as story illustrations and character icons, 'Stable Diffusion WebUI Forge' is used. By combining these tools, a mechanism is created where both text and visuals can be generated within your own PC.
- Text AI: Ollama (gemma4:12b-it-q8_0)
- Image AI: Stable Diffusion WebUI Forge
- How to launch: Simply double-click a batch file
Surprisingly Smooth Drafting and Operability
One of the convenient features of 'Narratable' is a mechanism that creates the foundation of a story in an instant. By simply entering one line of prompt, it drafts a complete set including the world-building, characters, glossary (Lorebook: specific setting materials for the work), and plot.
The time required for preparation is surprisingly short. For example, if there are 3 characters, the draft is completed in approximately 45 seconds. Furthermore, I paid close attention to operability; it is designed to reside in the system tray, so it can be called up immediately as long as Ollama is running.
- Preparation for 3 characters: Approx. 45 seconds
- PWA (Progressive Web App) support allows operation like an app from smartphones or tablets within the same Wi-Fi by adding it to the home screen
Design Considering PC Load
An important factor in running local AI is the management of PC memory (VRAM: memory on the video card). Narratable features a mechanism that automatically releases the model from VRAM upon closing.
Specifically, when using gemma4:12b-it-q8_0, it releases approximately 13.1GB of memory upon closing. Additionally, it is designed to save in-progress data before closing, ensuring that interrupting or resuming production can be done smoothly.
- Model release: Releases approx. 13.1GB for gemma4:12b-it-q8_0
- Prevents data loss through an auto-save function
- Manages everything from launch to exit with a single batch file
Next Preview
In this post, I discussed the overall picture and basic structure of the tool, but in the next installment (Part 2), I will explain the mechanisms in further depth.
The theme is a mechanism where an AI director decides 'who speaks.' I plan to share details about the refinements made to achieve more natural and complex ensemble dramas.
- Next theme: AI Director System
- Revealing the mechanism to increase story realism
- People who want to run AI on their own PC
- People who want to utilize AI in environments without an internet connection
- People who want to enjoy stories or roleplay using AI
- Ollama
- A tool for easily running various Large Language Models (LLMs) on your own PC.
- Stable Diffusion WebUI Forge
- One of the interfaces for operating the image generation AI 'Stable Diffusion' from a browser screen.
- VRAM (Video Memory)
- Dedicated memory mounted on a graphics board (GPU) used for image processing and AI calculations.
- PWA (Progressive Web App)
- Technology that allows you to install and operate a website like an app on a smartphone.
FAQ
How does it work to operate from a smartphone?
Because it supports PWA (Progressive Web App), as long as you are in the same Wi-Fi environment, you can access the PC as a server from your smartphone's home screen.
Why is releasing the model important?
Because AI models consume a very large amount of VRAM (Video Memory), releasing them when not in use ensures that other tasks and applications run smoothly.
'Narratable,' built by the operator of this account, is a tool for creating stories with multiple AIs in a completely local environment. It balances the convenience of constructing worlds from a single prompt and smartphone operability with a design that minimizes PC load.