AI controls
Last updated: October 26, 2026
AI controls decide what your team can do with AI in Layer: which models and kinds of generation they can use, how far the agent goes on its own, how much mature content model providers allow, and whether work can leave the workspace. You set them once for the whole studio and adjust them where a team, a project or one person needs something different.
How it works
AI controls can be set at five levels. Each level is more specific than the one before it:
Parent workspaces. In a studio with parent and child workspaces, every workspace above yours.
The workspace. Your workspace's own controls, for everyone in it.
The project. Controls for work done in one project, whoever does it.
Groups. Controls for the members of a group.
The member. Controls for one person.
Every control works the same way at every level. You choose Inherit, or you set a value (On or Off, a level, an amount). A control left on Inherit follows the levels above it. When several levels set the same control, the most specific one wins: a member's own setting beats their group's, a group's beats the project's, the project's beats the workspace's, and the workspace's beats its parent's. When no level sets a control, Layer's default applies, which is everything on.
Each model, each kind of generation and each part of the agent is a control of its own. A workspace can turn one video model off while a group turns a different one on, and both settings hold.
Locks
A workspace or a project can lock a value it sets. A locked value holds for every level below it: a child workspace, a project, a group or a member cannot change it, and their pages show it read-only, for example Locked by the Nightfall Studios workspace: Off. Use a lock for what must not vary, such as a model your publisher has not cleared. Leave a value unlocked when it is a sensible default that teams may change.
Groups and members cannot lock. Set a value before you lock it: the lock button stays unavailable on a control left on Inherit.
The levels
Parent workspaces
A child workspace inherits its parent's AI controls as they stand, and follows every later change. It can override anything the parent has not locked. See Manage AI controls across parent and child workspaces.
Workspace
Set under Settings, in the AI Controls pages, by anyone with Manage workspace settings. Available on every plan. See Set AI controls for a workspace.
Project
Set in the AI controls card of a project's settings, by anyone with Edit content policy. A project can lock a value for everyone who works in it, groups and members included. Enterprise only. See Set AI controls for a project.
Group
Set on the Model controls and Agent controls tabs of a group. When a member's groups disagree, the more permissive setting wins: a member in one group that turns a model off and another that turns it on can use it. Enterprise only. See Set AI controls for a group.
Member
Set from a member's menu on the Members page. A member's own controls take precedence over their groups, except where the workspace or a project locks a value. Enterprise only. See Set AI controls for one member.
Where each control can be set
Most controls can be set at every level. A few belong to where the work is made, or to the workspace alone:
| Control | Page | Levels |
|---|---|---|
| Models and kinds of generation | Models | All |
| Layer agent and its subagents | Agents | All |
| Spend confirmation | Agents | All |
| Content tolerance | Content Policy | Workspace, project |
| Input guardrails | Input Guardrails | Workspace, project |
| Output scoring | Output Scoring | Workspace, project |
| Reference set and workflow sharing | Sharing | Workspace |
A project can only make the content tolerance stricter than its workspace's, and it adds guardrails and scoring rules on top of the workspace's rather than replacing them.
The controls
Models
Which kinds of generation (image, video, 3D, audio) and which models your team can use, whether models nobody has turned on by name can run (including ones Layer adds later), and the settings each model's generations start from. A model that is off cannot be generated with or trained on. See Choose which models your team can use.
Agents
Whether the Layer agent and each of its subagents can be used, and the Creative Unit amount above which the agent asks before it runs a job. See Control the Layer agent and its spending.
Content tolerance
How permissive model providers are with mature content, from 1 (strictest) to 6 (most permissive). See Set your workspace's content tolerance.
Input guardrails and output scoring
Rules that screen a request before it generates, and rules that score what comes back. They are set as rules rather than switches. See Input guardrails and Set up scoring rules for a workspace or project.
Sharing
Whether the workspace's reference sets and workflows can be made public or linked into other workspaces. See Keep reference sets and workflows inside your workspace.
What it means for you
Check what someone actually gets. Preview as a member, on the Models and Agents pages, shows what one person ends up with across the workspace, their groups and their own controls, and which level set each value.
Every control says where its value comes from. Under each control you see what it inherits and from which level, for example Inherits Off from the Nightfall Studios workspace, or that your setting replaces it.
Each page saves as a whole. Changes collect until you select Save. Discard drops them, and Inherit everything puts every control on the page back on Inherit.
A permissive group can open a door. Because the more permissive group wins, turning a model on for one group gives it to everyone in that group, whatever their other groups say. Lock the value at the workspace when that must not happen.