Layer gives you two complementary controls over the quality and compliance of your generated assets: Input Guardrails screen requests before a model runs, and Output Scoring grades the results after generation. Both are configured per workspace in Settings.

Input Guardrails let you define what your workspace refuses to generate. Every request is checked against your active guardrails before it reaches the model — nothing is generated until the check passes (or, if you choose Warn mode, the request is flagged but allowed through).

Two guardrails are included out of the box:

Both built-in guardrails can be toggled between Block (the request is stopped) and Warn (the request proceeds but is flagged). You can also add Exceptions — for example, listing your own back-catalogue titles so they are always permitted even when the copyright rule would otherwise flag them.

Click + Custom rule (or + IP & trademark infringement for a dedicated IP template) to add your own guardrail. Give it a name, write a plain-language description of what to reject (as you would explain it to a reviewer), and set it to Block or Warn. You can also attach up to 10 reference images to help the model understand the rule.

Example custom rule: "Reject anything naming Clint Eastwood" — once saved, any prompt that references that name is blocked before generation starts.

When a request is blocked, the user sees a clear error screen showing the filename, the name of the guardrail that triggered, and the reason it was rejected. No generation credit is consumed for blocked requests.

Output Scoring automatically judges generated assets against one or more style standards you define. The score (0–100) appears on every session tile so your team can quickly see which outputs hit the mark.

In Settings → Output Scoring, click + Scoring rule to create a standard. Give the rule a name (e.g. "Fantasy") and write a description of the standard as you would to a new artist — for example: "Hand-painted fantasy style, visible brushwork." You can attach up to 10 reference images to sharpen the scoring.

Choose when the rule runs:

One rule can be marked as the Headline score; this is the score shown prominently on the session tile.

The score badge on a tile shows the headline score at a glance. Clicking through to an asset reveals a full breakdown per rule, including an explanation of why the asset received that score and suggestions for how to improve it — for example, which style elements were present and which were missing.

Use the Score filter on the board to surface assets by quality tier:

This makes it easy to find low-scoring assets for revision or to pull a shortlist of your best outputs.

If you update a scoring rule, existing assets keep their old scores. Use Re-score the library to apply your latest rules to assets already in the workspace. You can limit the run by asset age ("Assets to include") or set a cap on the number of assets processed ("Stop after"). Each asset costs one model call per rule, so narrow the run before starting a large one.

Guardrails are evaluated by a vision-capable AI model, which means they understand context and can evaluate image content as well as text prompts.

Yes. You can add as many scoring rules as you need and they run in parallel. The Headline score designation controls which rule's result is shown on the tile badge.

Input Guardrails and Output Scoring are available for all workspaces.