v0.1.0-preview.5

Windows preview 5

LoRA Pilot - WIndows Installer v0.1.0-preview.4

LoRA Pilot Windows Preview

This is a preview Windows release of LoRA Pilot delivered through a WSL2-backed installer.

It is meant to be testable, not invisible. Expect a few rough edges, especially around first-time WSL setup, SmartScreen, and GPU-specific behavior. The basic install/start path works. That is already more than many preview installers can say with a straight face.

What This Release Includes

• LoRAPilotSetup.exe
• windows-runtime-manifest.json
• runtime checksum files
• public runtime bundles hosted on Cloudflare R2 and downloaded during setup

You do not need to manually download the large runtime bundles when using a tagged preview release. The installer fetches them for you.

Requirements

• Windows 11 or Windows 10 22H2+
• Administrator rights
• WSL2 support enabled or installable on the machine
• roughly 100 GB free disk space if you want enough room for the runtime, models, and workspace data
• a decent internet connection, because the runtime bundle is not small and pretending otherwise would be unserious
• NVIDIA Windows drivers with WSL GPU support if you want GPU acceleration

Quick Start

1. Download LoRAPilotSetup.exe from the GitHub pre-release.
2. Run it as Administrator.
3. Let the installer:
   • check Windows support
   • enable or reuse WSL
   • download the runtime manifest and runtime bundle
   • import the managed LoRAPilot WSL distro
   • create Start Menu shortcuts
4. If Windows requires a reboot for WSL setup, reboot and sign in again.
5. Start LoRA Pilot from the Start Menu.

ControlPilot should open on http://127.0.0.1:7878 once the runtime becomes healthy.

What Gets Installed

• Windows launcher:
  %LOCALAPPDATA%\Programs\LoRAPilot\LoRAPilotLauncher.exe
• State, logs, downloads, and managed WSL files:
  %LOCALAPPDATA%\LoRAPilot
• Managed WSL distro name:
  LoRAPilot
• Distro storage location:
  %LOCALAPPDATA%\LoRAPilot\wsl\LoRAPilot

Persistent app data remains inside the distro under /workspace.

Start Menu Entries

• LoRA Pilot: starts the runtime and opens ControlPilot
• Open ControlPilot: opens the browser if ControlPilot is already running
• Stop LoRA Pilot: stops the managed runtime
• Uninstall LoRA Pilot: removes the Windows launcher and optionally the managed distro

Useful Commands

Open an elevated PowerShell and use:

$exe = "$env:LOCALAPPDATA\Programs\LoRAPilot\LoRAPilotLauncher.exe"

& $exe status --json
& $exe start
& $exe stop
& $exe open

That is much more informative than angrily clicking the same shortcut five times and expecting Windows to develop empathy.

Troubleshooting

Installer says WSL needs a reboot

• Reboot Windows.
• Sign back in.
• If setup does not resume automatically, run the installer again as Administrator.

Installer fails during the WSL import step

Clean stale state and retry:

wsl --unregister LoRAPilot
Remove-Item -Recurse -Force "$env:LOCALAPPDATA\LoRAPilot\wsl\LoRAPilot" -ErrorAction SilentlyContinue
Remove-Item -Force "$env:LOCALAPPDATA\LoRAPilot\state.json" -ErrorAction SilentlyContinue

Then rerun LoRAPilotSetup.exe.

ControlPilot does not open

Check launcher status:

$exe = "$env:LOCALAPPDATA\Programs\LoRAPilot\LoRAPilotLauncher.exe"
& $exe status --json

Check launcher logs:

• %LOCALAPPDATA%\LoRAPilot\logs\launcher.log

Windows warns that the installer is suspicious

That is still possible on preview builds, especially before signing and reputation are fully in place. If Windows blocks execution, use the usual SmartScreen “More info” path only if you trust the source of the release.

Port conflicts

LoRA Pilot expects these localhost ports:

• 7878
• 8888
• 8443
• 5555
• 6666
• 4444
• 9090
• 8675
• 7879

If 7878 is taken, for example:

netstat -ano | findstr :7878

Stop the conflicting process, then start LoRA Pilot again.

GPU is missing inside WSL

Run:

nvidia-smi
wsl --status
wsl --update

If the Windows NVIDIA driver is wrong, LoRA Pilot cannot compensate with optimism.

Current Limitations

• Preview quality, not GA
• No offline single-file installer
• Windows GPU validation is still thinner than it should be
• Some install/retry edge cases are still being tightened

If You Are Testing CI Artifacts Instead of a Tagged Preview Release

That is a separate, more annoying path.

Use:

• LoRAPilotSetup.exe
• Install-LoRAPilotPreview.ps1

The helper script exists for artifact-based smoke tests where the runtime is not published like a normal release yet. If you are using a tagged GitHub pre-release, you should not need it.

Lora Pilot v1.99

LoRA Pilot v1.99 Release Notes

This is last version of Lora-Pilot before v2.0. Lora-Pilot 2.0 will introduce some revolutionary features such as Copilot integration, LoRA testing and management, media browser, support for new models and much more!

These are the changes in v1.99 since the initial release:

Version 1.99

• Added ComfyUI-Downloader custom node to the image and workspace install path.
• ComfyUI user assets now mirror into the frontend static root to prevent 404s for user.css/userdata/templates.
• TrainPilot shows a real progress bar parsed from Kohya step logs (status no longer echoes raw log line).
• Headless TrainPilot stability fixes: corrected SRC_PATH/TRAIN_DIR logging and unbound variable crash.
• Models page redesign with card layout, inline filters, and click-to-copy model paths with truncation.
• New models added to models.manifest
• Docs section now renders markdown
• Changelog added to Docs

_New models page in ControlPilot_

Version 1.9

• ControlPilot shutdown scheduler defaults to safe stop for local storage; remove/delete requires explicit RUNPOD_POD_SHUTDOWN=remove or RUNPOD_VOLUME_TYPE=network.
• Jupyter allow-origin defaults to RunPod proxy + localhost/127.0.0.1; add more with JUPYTER_ALLOW_ORIGIN_PAT.
• ControlPilot model sizing/deletes fixed for hf_repo and nested hf_file paths.
• modelsgui now honors /workspace/config/models.manifest when present.
• Fixed ComfyUI proxy httpx import issue - moved to module-level imports for localhost compatibility.
• TrainPilot log reader now scans /workspace/outputs/*/_logs/train.log for Kohya output.
• Added universal deployment support for ComfyUI live preview (RunPod + localhost).
• Fixed mixed content issues in ComfyUI integration for both environments.
• Major architectural refactor: Extracted services into modular structure (comfy.py, models.py, shutdown.py).
• Added comprehensive Docker Compose setup for local development with GPU/CPU/Dev modes.
• Added docker-compose-helper.sh script for easy container management.
• Added environment variable configuration (.env.example) for customization.
• Added health checks for local deployment and documented Compose resource-limit caveats.

Version 1.8

• updated InvokeAI to 6.10.0
• several UX fixes and improvements in Control Pilot
• addressed several critical security issues
• added shutdown scheduler to ControlPilot (schedule shutdown in X minutes/hours/days in Dashboard); supports pod with local storage, network volumes and local docker images. In order for shutdown schedule to work with pods using network volumes, env variable RUNPOD_VOLUME_TYPE must be set to "network" (RUNPOD_VOLUME_TYPE=network).

Added "Schedule a shutdown" feature to dashboard.

!!!WARNING!!! Setting RUNPOD_VOLUME_TYPE to "network" will cause the pod to terminate (delete) when the shutdown schedule is triggered. Use only with pods with mounted network volumes.