AI Workflow Hub – Deploy Local AI in Minutes

Available Workflow Packages

Flux2 Image Generation

The Flux2 workflow wraps the FLUX.2 model family into a turnkey local service. It supports resolutions from 512×512 to 2048×2048, configurable step counts, and LoRA adapter loading. The package includes an MCP server that exposes generation, parameter adjustment, and image retrieval as callable tools for AI-agent integration. Minimum system memory is 16 GB; recommended VRAM is 24 GB or higher. Installation takes approximately three minutes on a modern Linux workstation with an NVIDIA GPU.

WhisperX Transcription Pipeline

This workflow packages WhisperX v3.8.6 with optimized ONNX Runtime execution for local speech transcription. It handles files up to two hours in duration, supports 97 languages, and outputs SRT, VTT, JSON, and plain-text formats. The pipeline runs entirely on GPU with batched inference for throughput up to 30× realtime on an RTX 4090. No audio data leaves the local machine during processing.

LLM Serving Stack

The LLM serving workflow bundles llama.cpp v0.4.0 with a REST API layer and WebSocket streaming support. It loads models in GGUF format, manages KV-cache allocation, and exposes OpenAI-compatible endpoints for drop-in replacement of cloud API calls. Quantization profiles from Q4_K_M through Q8_0 are pre-configured. The package supports concurrent requests with automatic batching and includes a health-check endpoint for container-orchestration environments.

Real-ESRGAN Upscaling

Image upscaling workflow using Real-ESRGAN with tile-based processing to handle arbitrarily large inputs. Output resolutions up to 8K are supported on hardware with 8 GB VRAM. The package includes batch-processing mode for directories and an MCP tool interface for automated integration into media pipelines.

Installation Procedure

1. Download the workflow archive from the Homegrown Intelligence storefront.
2. Extract to a target directory: tar xzf workflow-flux2-2.1.0.tar.gz
3. Run the setup script: ./setup.sh – this verifies GPU drivers, pulls the model weights, and creates a systemd service file.
4. Start the service: systemctl --user start hgi-flux2
5. Verify the endpoint responds: curl http://localhost:8080/health

Detailed troubleshooting for driver mismatches, port conflicts, and out-of-memory conditions is covered in the support knowledge base.

System Requirements