# train-lora

train-lora is a Claude Code skill that validates LoRA training with Draw Things through draw-things-cli. It runs local dataset checks and compares generated images with and without the LoRA.

- skill · `type=skill` · Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf (Devin Desktop), Claude.ai · ★ 171 repo stars
- Author: drawthingsai
- License: GPL-3.0 · policy=link-only · description only; the file stays with the author
- Source: https://github.com/drawthingsai/local-code-community/tree/HEAD/.agents/skills/train-lora/SKILL.md
- Updated: 2026-11-17
- Security check: automated, 2026-11-17 (text) — no signs of malicious code found.
- HTML: https://affpapa.org/en/nevibecode/skills/train-lora

## What it does

train-lora checks LoRA training from a short trial run through image comparison. First, build the optimized CLI with `bazel build --compilation_mode=opt //Apps:DrawThingsCLI`, then run `bazel-bin/Apps/DrawThingsCLI` for the training and generation commands.

A single-image test uses a local dataset folder containing the image and a matching `.txt` caption file. The workflow checks runs at 1, 20, 100, and 500 steps, tracking loss, scale, and checkpoint output. For generation checks, it holds the prompt, seed, and settings constant between the base model and the LoRA. Graph compilation in optimized mode can take 5 to 30 minutes on heavy trainers.

## Install

### Claude Code (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add drawthingsai/local-code-community --skill train-lora -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/train-lora/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/train-lora/

### Codex

1. **Skills CLI (Vercel)**
```bash
npx skills add drawthingsai/local-code-community --skill train-lora -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/train-lora/SKILL.md` — For all projects use ~/.agents/skills/

### Cursor

1. **Skills CLI (Vercel)**
```bash
npx skills add drawthingsai/local-code-community --skill train-lora -a cursor
```

2. **Manually: put the files in place**
- File: `.cursor/skills/train-lora/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install drawthingsai/local-code-community .agents/skills/train-lora
```

2. **Manually: put the files in place**
- File: `.github/skills/train-lora/SKILL.md`

### Gemini CLI

1. **Skills CLI (Vercel)**
```bash
npx skills add drawthingsai/local-code-community --skill train-lora -a gemini-cli
```

2. **Platform CLI**
```bash
gemini skills install https://github.com/drawthingsai/local-code-community --path .agents/skills/train-lora
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/train-lora/SKILL.md` — The legacy .windsurf/skills/ path still works

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/drawthingsai/local-code-community/tree/HEAD/.agents/skills/train-lora/SKILL.md

## When to use it
- check graph compilation with a short training run
- track loss and scale during training
- verify that a checkpoint contains nonzero LoRA weights
- compare base and LoRA images under matching settings

## Example prompts
- Validate Draw Things LoRA training with a small local dataset.
- Check loss, scale, and checkpoint output during a LoRA run.
- Compare base-model and LoRA images using identical generation settings.

## Similar
- [lora-qlora-recipes](https://affpapa.org/en/nevibecode/skills/lora-qlora-recipes.md) — skill · 2600 installs
- [huggingface-lora-space-builder](https://affpapa.org/en/nevibecode/skills/huggingface-lora-space-builder.md) — skill · 963 installs
- [anymove-lora](https://affpapa.org/en/nevibecode/skills/anymove-lora.md) — skill · ★ 653 repo stars
- [fal-train](https://affpapa.org/en/nevibecode/skills/fal-train.md) — skill · 2900 installs
- [train-sentence-transformers](https://affpapa.org/en/nevibecode/skills/train-sentence-transformers.md) — skill · 1700 installs
- [tao-train-mask-auto-encoder](https://affpapa.org/en/nevibecode/skills/tao-train-mask-auto-encoder.md) — skill · 1600 installs

From the same set: [drawthingsai/local-code-community](https://affpapa.org/en/nevibecode/by/drawthingsai/local-code-community.md) — 20 cards

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md
