VRGDG Krea 2 AI Toolkit Installer

VRGDG Krea 2 AI Toolkit Installer sets up Ostris' AI Toolkit - the community's default LoRA trainer - in an isolated virtual environment, specifically for the pack's experimental paired Krea 2 Edit LoRA training. If the Musubi installer is the pack's main Krea 2 path, this is the alternate route it offers for a different flavor of training, and it's honest about being experimental: the tooltip says so, and you should read that as "unstable, iterate carefully."

For context on why AI Toolkit: it's the one-person training suite that became the default answer to "how do I train a LoRA on this?" for every new architecture - it was training Flux a day after release, Z-Image Turbo two days after, and it's the harness most third-party training research builds on. The flip side, which the KB documents from real user reports, is that the install can be rough ("half the dependencies wouldn't want to load"). That's exactly why this node exists: it automates the painful install into a clean, isolated venv rather than letting AI Toolkit's dependencies fight your ComfyUI environment.

The input

• target_root - where the isolated AI Toolkit environment should be created (default A:/MUSUBI). The venv is created under here, separate from your ComfyUI Python.

The outputs

• install_root - where the toolkit landed.
• python_path - the path to the venv's Python, which is what the training wrapper will invoke.
• status - human-readable state.
• report_path - the install log for debugging.

Installing it

It's a node in the VRGameDevGirl pack, so: Manager → search vrgamedev, or git clone https://github.com/vrgamegirl19/comfyui-vrgamedevgirl.git into custom_nodes/, restart, hard-refresh. Then run this node once and let it do its thing - the whole point is that you don't hand-install AI Toolkit.

Common issues

• The install takes a long time. AI Toolkit's dependency set is large (and the KB notes it can be genuinely painful). Give it time, then read report_path if something fails.
• Training still doesn't see the venv. The python_path output is there for a reason - the training wrapper needs to use that Python, not your system Python. Check the wiring.
• It's experimental, so it breaks. Krea 2 Edit paired training is early days. If a run fails mid-way, the report/log outputs are your friend - don't assume the node is broken just because the training technique is bleeding-edge.

When would you pick this over the Musubi installer? If your training style leans on AI Toolkit's GUI/config approach or you're specifically chasing the Edit-LoRA pairing. If you just want a Krea 2 LoRA trained, the Musubi path is the more proven one. This node is the "different tool, same goal" option.