🤮 AI Slop Manager Final v001 - Automatised Agentic ComfyUI workflows

https://youtu.be/EjBjWNgTJnU?si=Ah4OgVj0EJKnI2bQ

🤮 What is "AI Slop Manager"?

AI Slop Manager is not just another wrapper—it is a heavy-duty, locally hosted orchestration engine built for serious AI production.

Designed to sit on top of ComfyUI and Ollama, it replaces the chaotic "spaghetti node" interfaces with a sleek, modular Gradio 4 dashboard. Whether you are generating studio-quality character sheets, bulk-producing 3D assets via Trellis, or rendering cinematic Wan 2.2 videos, this software acts as the "Director." It forces local Large Language Models (LLMs) to write mathematically precise prompts and dynamically manages server hardware to ensure your GPU (the "Muscle Machine") never runs out of memory.

⚙️ How It Works: The Architecture

At its core, the software is built on stability, modularity, and aggressive error handling.

• The Frontend (Gradio 4 with Anti-Crash Armor): React-based UIs often crash when hiding large dynamic components. We engineered a proprietary CSS-based visibility hack (hide-slop) that eliminates "Phantom Clicks" and "Ghost Checkbox" crashes, keeping the UI lightning-fast and bulletproof.

• The Backend (Python 3 & ComfyUI API): Communicates directly with ComfyUI via WebSockets, dynamically injecting parameters, 32-bit safe seeds, and complex multi-view image nodes directly into JSON workflows.

• VRAM Management A standalone Flask REST API (port 9000) that acts as a ruthless bouncer for your GPU. You never hit an Out Of Memory (OOM) error because the VRAM is flushed before switching tasks.

• File & Metadata Management: Features an OpenCV and PIL-powered gallery that reads the exact physical resolutions of generated MP4s, GIFs, and PNGs, alongside a metadata extractor that logs every prompt, seed, and workflow used.

• 
  

🚀 The Arsenal: Completed Features

We have successfully implemented a robust suite of production-ready features:

• Smart Server Selector: Instant switching between Image, Video, and 3D ComfyUI servers via remote .bat execution, with safety protocols against Windows Subprocess pathing errors.

• Context-Aware Dynamic UI: The interface shape-shifts based on the selected workflow. It intelligently hides batch counts during 3D generations to prevent crashes, reveals specific sizing modes (e.g., Resize, Stretch, Pad) only for Video models, and features 1-click aspect ratio presets.

• Ollama Persona Engine: Hardcoded, guardrailed prompt engineering. You type "a warrior," and Ollama injects strict anatomy protections (preventing bad hands/faces) and formats it into specific styles like 🎮 Game Optimized 3D or 🎬 Cinematic Action Slow-Mo.

• Auto 3D Pipeline (Trellis): A seamless chain that takes an image, strips the background, and pushes it through the Trellis 3D model, complete with a strict 32-bit seed limit wrapper to prevent native ComfyUI math crashes.

• Dynamic Preset Memory: A completely automated "Save State" system that zips all UI parameters using Python *args, ensuring no configurations are lost between sessions.

• 
  

🗺️ Development Roadmap: What's Next?

The engine is running, but we are expanding the factory. The upcoming pipeline includes:

1. Local RAG (Retrieval-Augmented Generation) Knowledge Base: Integrating ChromaDB with Ollama's embedding models. This will allow the user to "chat" with their own database of downloaded YouTube transcripts, GitHub repos, and JSON workflows to get instant, hallucination-free advice on node building.

2. Automated Model-Archiver (Garbage Collector): 

3. Advanced Video Integration: 

4. Storyboard Mode: A sequential generation tool for comic books and continuous narrative videos.

5. Smooth asset Integration to Houdini/Unreal