Vellium: Control Story Mood with Sliders, Not Prompts

What It Is

Vellium is a desktop application that replaces traditional prompt engineering with visual slider controls for AI-generated storytelling. Instead of manually tweaking prompts to shift a scene from tense to relaxed or adjust dialogue formality, writers manipulate eight dedicated sliders that handle these adjustments automatically. The tool operates in two modes: chat mode for character conversations and writer mode for longer narrative projects where each chapter can maintain distinct atmospheric settings.

The application connects to various language model backends including Ollama, LM Studio, OpenAI, and OpenRouter. Characters created in chat mode transfer directly to writing projects, allowing writers to develop personalities through conversation before deploying them in structured narratives. Vellium imports SillyTavern V2 character cards and retrieves avatar images from Chub, making it compatible with existing character libraries many fiction writers already maintain.

Why It Matters

Traditional prompt engineering creates friction in creative workflows. Writers interrupt their narrative flow to edit system prompts, test variations, and troubleshoot when the model’s tone drifts. This technical overhead pulls focus from storytelling to prompt debugging. Vellium addresses this by abstracting prompt construction into intuitive controls - adjusting “Pacing” or “Intensity” becomes as simple as mixing audio levels.

The character portability between modes solves a common problem in AI-assisted fiction: inconsistency. Developing a character through conversational exchanges before inserting them into formal narrative helps establish voice and personality traits. Writers can interview their characters, test dialogue patterns, and refine quirks before committing them to the main story structure.

For developers building narrative tools, Vellium demonstrates an alternative interface paradigm. Most AI writing assistants still rely on text-based configuration. Slider-based controls suggest possibilities for more granular, real-time narrative control without requiring users to learn prompt syntax. The inclusion of lorebooks, MCP tool calling, and multi-agent chat indicates ambitions beyond simple mood adjustment toward comprehensive worldbuilding infrastructure.

Getting Started

The project lives at https://github.com/tg-prplx/vellium with installation instructions in the repository. After cloning and setting up the application, writers need to configure a model backend. For local inference with Ollama:

Point Vellium to the Ollama endpoint (typically http://localhost:11434) in the settings. The eight sliders in chat mode control Mood, Intensity, Dialogue Style, Pacing, Description Detail, Emotional Depth, Narrative Voice, and Scene Focus. Experimenting with combinations reveals how these parameters interact - high Intensity with slow Pacing creates simmering tension, while casual Dialogue Style with high Emotional Depth produces intimate conversations.

To import existing characters, drop SillyTavern V2 JSON files into the appropriate directory. The application fetches associated avatars from Chub automatically if character definitions include the proper identifiers.

Context

Vellium occupies a niche between general-purpose chatbots and specialized writing software. Tools like NovelAI offer extensive narrative controls but operate as cloud services with subscription models. SillyTavern provides deep customization through prompt templates but requires significant technical knowledge. Vellium attempts a middle path: local-first operation with simplified controls.

The slider approach has limitations. Complex tonal shifts might require more nuance than eight parameters provide. Writers accustomed to precise prompt control may find the abstraction restrictive. The project remains in active development, meaning features and stability will evolve.

Alternatives include KoboldAI for local inference with web-based interfaces, or LM Studio’s built-in chat for simpler conversational use cases. Writers prioritizing cloud-based solutions might consider Sudowrite or Claude with custom system prompts. Each tool makes different tradeoffs between control granularity, ease of use, and infrastructure requirements.

The multi-agent chat and MCP tool calling features suggest Vellium aims beyond solo writing toward collaborative storytelling or complex narrative simulations. Whether these advanced features integrate smoothly with the core slider interface remains to be seen as the project matures.