Custom ComfyUI workflow for LLM based local tarot card readings!
This custom workflow draws cards, interprets them with LLM logic, and generates unique art influenced by your questions.
A developer known as DoctorDiffusion has created ProtoTeller, an innovative custom workflow for the popular node-based AI interface, ComfyUI. Unlike standard node packs, ProtoTeller is designed as an interactive experience, blending generative AI with the mystique of tarot. The system employs a custom wildcard solution to simulate drawing cards from a deck. It then chains prompts to a local Large Language Model (LLM), which generates a unique interpretation for each card. A key feature is its handling of reversed cards, where the LLM logic adapts to provide a different, often opposing, reading.
The workflow is highly customizable and user-influenced. Users begin by entering a topic—such as 'Financial Future'—or a direct question. This input guides both the visual generation of the card art and the textual narrative created by the LLM. A second input allows for style keywords or custom LoRA (Low-Rank Adaptation) tokens, letting users tailor the aesthetic without requiring the creator's specific models. For organization and ease of use, the entire process is packaged inside a ComfyUI subgraph, and every output—an image and a corresponding .txt file of the reading—is automatically saved. Currently available on GitHub as ComfyUI-ProtoTeller, the project represents a creative fusion of stable diffusion and LLM capabilities, moving beyond pure image generation into structured, interactive storytelling.
- Uses a custom wildcard system and chained LLM prompting to generate unique tarot card readings and corresponding art.
- Interprets card orientation, with reversed/inverted cards triggering different LLM logic for altered readings.
- Accepts user topics/questions and style/LoRA inputs to personalize both the visual output and textual interpretation.
Why It Matters
Demonstrates a novel, practical application of local AI, combining image generation and LLMs for personalized, interactive entertainment and creative exploration.