The ultimate multi-modal local character engine and automated visual pipeline. Seamlessly link SillyTavern and ComfyUI to host high-intelligence local LLMs alongside heavy Diffusion pipelines, generating rich contextual images in real-time based on combined text dialogue and primitive concept sketches!

This repository is distributed under the SUPERKLOK LABS UNIFIED PUBLIC ASSET LICENSE v1.0:

100% Free for Commercial Use: You are fully permitted to use this workspace, automation workflows, scripts, and visual pipelines to generate commercial artwork.

Mandatory Credit: You must credit Superklok Labs / TrenchPetals by linking to or tagging one of our official handles (Twitter(X) or Instagram) whenever you publish content created or upscaled with this asset.

Unlike standard character bots tied to restrictive cloud platform subscription fees or non-commercial licenses, this entire architecture is 100% free for commercial local deployment.

Indie Game Developers: Rapidly build interactive deep lore, complex dialogue trees, and production-ready concept art locally.

Content Creators: Automate dynamic, multi-modal livestream interactions and safe visual storytelling pipelines in real-time.

IP Creators: Turn primitive sketches into highly detailed master assets, retaining absolute ownership and monetization rights.

Dual-Engine VRAM Balancing: Intelligently shares a single 16GB VRAM graphics card between a heavy LLM and a raw Diffusion workspace without crashing drivers.

Bleeding-Edge Text & Logic Layer: Powered by Gemma-3-12B-IT (Q8_0) hosted via KoboldCPP into SillyTavern with an engineered 8,192-token context window for rock-solid short-term memory.

Advanced Multi-Modal Generation: Native conditioning on Stable Diffusion 3.5 Large (FP8 scaled) passing text prompts and custom canvas drawings directly through automated ControlNet and Img2Img nodes.

Granular Detailer Hooks: Pre-configured FaceDetailer, HandDetailer, and FootDetailer pipelines utilizing custom scheduling to eliminate anatomy mutations while preserving sketch fidelity.

Low-VRAM Compression: Advanced quantization text encoder paths drop local VRAM footprints by over 40% while preserving hyper-detailed 1344x768 outputs.

Definitive Pipeline Workflows: Pre-configured .json script arrays for both Multi-Colour Portrait (alien world generation) and Purple Portrait (deep cosmic brick sewer layout) systems.

Benchmark Core Files: Pre-formatted Character Card - Sunny Chuxem.json API profiles hardcoded to anchor automation variables (%prompt% and %seed%) flawlessly out of the box.

Production Canvas Assets: Highly optimized layout guides (ColourCanvas.png and StudioCanvas.png) to directly dictate scene vibrance and composition density.

Local Python Tooling: Dedicated standalone scripts including custom image captioning setups.

Built entirely for processing efficiency on local desktop hardware.

Operating System: Windows PC.

Hardware: NVIDIA GPU (12GB VRAM minimum, 16GB+ recommended for dual-engine workflows).

VRAM Compression: Runs an optimized FP8 startup batch execution (--lowvram --fp8_e4m3fn-text-enc), saving over 40% in VRAM overhead.

100% Portable Port: Operates entirely offline without cloud subscription fees or restrictive corporate filtering.

If you need custom-engineered generation pipelines, bespoke brand LoRAs, or enterprise-optimized local ComfyUI environments, let's build it together:

💼 Hire on Contra

🌐 Portfolio on GitHub

Due to the immediate, digital download nature of these files, workflows, and LoRA models, all sales are final and non-refundable. Once you gain access to the download package, we cannot revoke access, making it impossible to process returns. Please review the system hardware requirements carefully before completing your purchase.