Local LLM Server Manager

Appearance

Stable Diffusion Forge Integration ​

Stable Diffusion WebUI Forge is an optimized inference environment for image generation. Forge supports SD 1.5, SDXL, and FLUX.1 models with lower VRAM overhead than standard WebUI implementations.

Local LLM Server Manager provides lifecycle management, process containment via Win32 Job Objects, and direct CivitAI model downloads for Forge.

Engine Overview ​

  • Default Endpoint: http://127.0.0.1:7860
  • Health Check URL: GET http://127.0.0.1:7860/sdapi/v1/progress
  • Default Launch Script: webui-user.bat (Windows) or webui.sh (Linux)
  • Supported Architectures: SD 1.5, SDXL, Pony, FLUX.1 (Schnell / Dev), SD 3.5

NOTE

SD Forge runs on port 7860 by default. If you use a custom port, update the endpoint URL in the Settings tab.

Configuring Model Directories ​

Forge requires specific directory structures for diffusion checkpoints, LoRAs, and VAE files.

Forge Root /
├── models/
│   ├── Stable-diffusion/   <-- Checkpoint models (.safetensors)
│   ├── Lora/               <-- LoRA style weights (.safetensors)
│   └── VAE/                <-- Variational autoencoders (.safetensors)
└── webui-user.bat

Setting the Directory Path ​

  1. Open the Local LLM Server Manager dashboard.
  2. Navigate to the Stable Diffusion tab or open the Settings tab.
  3. Locate the Forge / SD Models Directory input field.
  4. Enter the absolute path to your Forge models folder (for example: C:\AI\SD_Forge\models).
  5. Locate the Forge Executable Path field.
  6. Enter the path to your launch script (for example: C:\AI\SD_Forge\webui-user.bat).
  7. Click Save Path. The manager writes your configuration to settings.json.

TIP

Use the Auto-Detect Installed Tools button on the Settings tab. The manager scans all connected drives and populates your paths automatically.

Booting and Stopping Forge Cleanly ​

Python machine learning applications spawn multiple sub-processes. Terminating a parent batch script often leaves orphaned Python processes running in the background. These orphaned processes hold GPU VRAM hostage.

Clean Termination with Win32 Job Objects ​

Local LLM Server Manager uses Windows Win32 Job Objects to guarantee complete process cleanup:

  1. When you click Boot SD Forge, the manager creates the process inside a monitored Win32 Job Object.
  2. Windows binds all subsequent child processes (Python runtimes, PyTorch CUDA workers, and API listeners) to that job object.
  3. When you click Stop SD Forge or exit the manager, the operating system terminates every process in the tree instantly.
  4. Your graphics card releases all allocated VRAM immediately without requiring a system reboot.

How to Start and Stop the Engine ​

  • To Start Forge: Navigate to the Stable Diffusion tab. Click Boot SD Forge. Check the terminal output window or wait for the health badge to turn green.
  • To Stop Forge: Click Stop SD Forge. The manager terminates the job object and clears the status indicator.

IMPORTANT

The VRAM Orchestrator communicates with Forge before launching heavy jobs. The orchestrator unloads active text models from Ollama automatically when Forge starts.

Downloading Models from CivitAI ​

You can search and download community checkpoints and LoRA styling weights directly from CivitAI without leaving the dashboard.

Step-by-Step Download Procedure ​

  1. Open the Stable Diffusion tab in the manager dashboard.
  2. Select the CivitAI search view.
  3. Type your search query into the search bar (for example: DreamShaper, Juggernaut XL, or Detail Tweaker).
  4. Select a category from the Model Type dropdown:
    • Checkpoint: Full base model weights.
    • LoRA: Lightweight styling and character adapters.
    • VAE: Color correction and latent decoders.
  5. Review the Can I Run It badge on each result card:
    • Full VRAM: Model fits your hardware with optimal generation speed.
    • Partial Offload: Model requires CPU memory offloading.
    • Out of Memory: Model size exceeds total available VRAM.
  6. Click your preferred model card to open the version selector.
  7. Click ⬇ Download to Forge.
  8. The manager downloads the .safetensors file directly to the appropriate folder (models/Stable-diffusion or models/Lora).
  9. Monitor the download progress bar on the dashboard.

WARNING

Full diffusion checkpoints (especially SDXL and FLUX) range from 6 GB to 24 GB in file size. Confirm that your target storage drive has adequate free space before starting large downloads.

Released under the MIT License.

Copyright © Spelech