# Curate and Apply Layer Reference Sets

Inconsistent datasets and incompatible models can weaken a trained style while still consuming Creative Units. This skill guides curation, training choices, weight tuning, and contribution checks before generation.

## Install

```bash
npx skillstore add layerai/layer-reference-sets
```

## Metadata

- Status: approved
- Slug: layerai-layer-reference-sets
- Skillstore revision: r1
- Version status: missing
- Tree hash: 68a7948968ae926f1ea45f38aee0d6a2804ed5722a09940b75eb9c30410369fc
- Author: layerai
- GitHub username: layerai
- License: MIT
- Repository: https://github.com/layerai/skills/tree/315d06db6f760160b26293fd867ac759942e2106/skills/layer-reference-sets
- Ref: ee0fcd6d9de455c669ab56b791cc7ead7b94348e
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands
- Quality score: 77
- Quality tier: bronze
- Public page: https://skillstore.pages.dev/skills/layerai-layer-reference-sets
- Manifest: https://skillstore.pages.dev/api/skills/layerai-layer-reference-sets/manifest

## Capabilities

- Recommend training or an approved reference image based on reuse, subject count, and project scope.
- Guide dataset curation for consistent rendering, appropriate variation, clear subjects, and clean images.
- Choose among character, object, scene, effect, and general training categories.
- Guide reference set weights from 0 to 2 and explain when weight does not apply.
- Diagnose degraded reference sets using contribution reports and model compatibility guidance.
- Outline training and generation workflows with price estimates, confirmation above 20 Creative Units, and status polling.

## Use Cases

- Maintain a studio style: Select consistent artwork, choose the general category, and plan a reusable reference set for ongoing team generation.
- Build a character library: Separate subjects into reference sets and vary poses, angles, and expressions while preserving each character design.
- Investigate weak style results: Review contribution warnings, model selection, and weight settings before spending on another generation.

## Prompt Templates

### Choose whether to train

```
I need [asset count] assets for [project duration] using [approved artwork]. Recommend a reference image or trained reference set, and explain the choice.
```

### Curate a training set

```
Review [supplied images] for [subject or style]. Recommend exclusions, useful variation, separate sets, and the appropriate training category. Do not upload anything yet.
```

### Plan a controlled training workflow

```
Plan training for [approved dataset]. Estimate costs before paid actions and request confirmation above 20 Creative Units. Check contributions before generation and report missing tools.
```

### Diagnose degraded style application

```
Analyze [contribution report] and [run settings] for [reference sets]. Explain degradation, model compatibility, and weight effects. Recommend a revised estimate before any new spend.
```

## Limitations

- This package provides guidance, not bundled scripts or a standalone training engine.
- Execution requires access to Layer tools; detailed training mechanics are delegated to the sibling layer-train-custom-style skill.
- Training and generation consume Creative Units and require time; the skill does not guarantee consistent results.
- Weight affects LoRA application only and is ignored when sets contribute reference images or prompt text.

## Best Practices

- Curate consistent, clean artwork and vary only the attributes that should change.
- Match each set to one subject or style and select its training category deliberately.
- Inspect price estimates and contribution reports before spending, and confirm actions above 20 Creative Units.

## Anti Patterns

- Pad a dataset with inconsistent images or mix unrelated subjects into one reference set.
- Pin a model without checking reference set compatibility, then treat weak output as a weight problem.
- Treat weight as a binary setting or claim a degraded set applied successfully.

## Security Audit

- Audited at: 2026-10-11T14:34:19.878\+00:00
- Summary: All 24 static findings are false positives: Markdown tool names, category labels, a user-requested installation example, and model-selection guidance. No evidence found of shell backtick execution, system reconnaissance, prompt injection, or unauthorized data transfer. The documented workflow uses Layer uploads and paid training or generation, with estimates and confirmation above 20 Creative Units.

## Stats

- Views: 0
- Downloads: 1
- Favorites: 0
- Popularity score: 0