VRGDG LoRA Dataset Creator UI

VRGDG LoRA Dataset Creator UI is the pack's built-in dataset workshop. It opens a standalone UI for building and reviewing captioned image datasets - the raw material for every LoRA you train through this pack, whether that's a character, a style, or the Krea 2 edit-pair experiments the training nodes support. It's a launcher node: no inputs to configure, one status output telling you to go open the UI.

Here's the thing about LoRA training nobody puts on the box: the model architecture has changed, but the dataset is still everything. The KB's LoRA training material hammers the same point that's been true since SD 1.5 - curation beats every training knob. Twenty well-captioned, consistent images will out-train sixty that were scraped together. This tool exists to make that curation step comfortable instead of a folder-management chore: you can review pairs, check captions, and catch the mislabeled images that quietly wreck a run.

What you'd use it for

• Building a character or style dataset: collect images, caption each, review before you feed it to a trainer.
• The Krea 2 / MiniMax / LTX training workflows this pack ships - they all want a dataset folder with image + .txt caption pairs, which is exactly the shape this UI produces.
• Catch problems early: wrong captions, duplicates, low-quality frames. A caption that says "no glasses" on a photo with glasses is a training bug you'll spend hours chasing later.

The output

• status - a string confirming the UI was launched. Nothing else to wire; the real work happens in the window it opens.

Installing it

Same as the whole pack - Manager, search vrgamedev, or:

cd ComfyUI/custom_nodes
git clone https://github.com/vrgamegirl19/comfyui-vrgamedevgirl.git

Restart ComfyUI and hard-refresh the browser - a launcher node's entire UI lives in the frontend, so a stale tab can leave the button dead. The pack's heavy pip requirements apply to the whole collection; on Windows portable, follow the README's pip tooling commands and prefer Python 3.12.

Common issues

• Clicking it does nothing. Stale frontend almost always. Hard-refresh and try again.
• The UI opens but your dataset isn't there. Check where the UI is pointing - it manages a specific folder structure, and you'll want your images/captions inside it before training.

If you're already comfortable curating datasets by hand, you can skip this node and use any folder tool. But if you're new to LoRA training and this pack is your entry point, use the thing the author built - it's a lot less lonely than a raw filesystem.