# Models

> Models are the engines behind what Endless makes — how to think about picking one, and what decides which ones you can use.

Source: https://endless.app/docs/concepts/models

A **model** is the engine that actually makes something: the image model that renders your
prompt, the video model that turns a scene into a clip, the chat model that writes the copy.
Endless doesn't have one — it has a catalog, and you pick the one that suits the job.

<Callout type="warn" icon="<Compass />">
  **This page deliberately doesn't list the models.** A model is added, repriced
  or retired without a new version of the app, so any list written here would
  start going stale the day it was published. Your live list is the model picker
  wherever you generate. People who can view workspace governance — admins, by
  default — also get the whole catalog with its costs under **Settings →
  Governance → Models**.
</Callout>

## How to choose one

You don't need to know how a model works to pick well. Three things separate them:

<Cards>
  <Card icon="<Photo />" title="What it makes">
    Some models make images, some make video, some make audio, some write. The
    picker only ever offers you models that fit what you're doing.
  </Card>

  <Card icon="<Star />" title="What it's good at">
    Within a kind, models differ — one is better at text inside an image,
    another at photographic realism, another at editing a picture you already
    have. The picker shows each model's name and provider.
  </Card>

  <Card icon="<Coins />" title="What it costs">
    Costs range from a credit or two to several hundred for one run. This is the
    difference that actually bites, so check it before you commit.
  </Card>
</Cards>

Each model's controls come from the model itself, so switching model changes the settings
available to you. That's expected — a model that supports a reference image offers one, and a
model that doesn't, doesn't.

## Different models, very different prices

The catalog spans a wide range. The composer shows the cost of the exact run you've set up
before you send it, and the governance page — if you can open it — lists every model's cost per
run in credits.

<Callout type="info">
  Some models don't have a fixed price and show &#x2A;*"Varies"*&#x2A; instead. The
  governance page explains why: &#x2A;"This model is priced by the size or length of
  its inputs, so it has no fixed per-output rate. The exact cost is calculated
  from the real inputs at each run."* Feeding it a longer clip or a bigger image
  costs more. The composer still shows you the real number once your inputs are
  in place.
</Callout>

<Cards>
  <Card icon="<Coins />" title="Credits and spending" href="/docs/concepts/credits-and-spending">
    When a generation is charged, and how close the estimate you see is to the
    amount you pay.
  </Card>
</Cards>

## Why you might not be able to use a model

Two different things get confused here, and the picker tells them apart for you: a model you
can **see but not pick**, and a model that **isn't there at all**.

### You can see it, you just can't pick it

Some models need a paid plan. Those stay in the picker — dimmed, unselectable, with a lock and
the name of the plan they need. Selecting one takes you to billing instead of switching model,
so nothing is generated and nothing is charged. Your plan never hides a model from the picker;
it only tells you what it would take to run it.

### It isn't in the list at all

A model the workspace turned off genuinely disappears — from the pickers, from `@&#x60; mentions and
from what the chat can reach — and the block is enforced when the run is requested, not just in
the interface. Anyone who can change workspace governance settings can restrict the catalog,
described in the app as: &#x2A;"Choose which AI models are available, for the whole workspace, a
team, or one person. Turning a model off hides it from every generator and stops new spend."*

Whole surfaces can go the same way. Image generation, video generation and the model picker
itself are features a workspace can turn off entirely. If the surface is missing rather than
the model, that's why.

### How workspace restrictions stack

Restrictions are set at three scopes — **Workspace**, **Team** and **Person*&#x2A; — and they only
ever narrow. Of the workspace scope the app says &#x2A;"The default for everyone in the
workspace."&#x2A; Of the two narrower ones: &#x2A;"Rules on top of the workspace default. Anything turned
off for the workspace stays off here."*

<Callout icon="<Lock />">
  You can't grant back at a narrower scope what the workspace took away. A model
  off for the workspace is off for every team and every person in it.
</Callout>

## Two catalogs, one page

For the people who can open it, **Settings → Governance → Models** has a **Media** tab and a
**Chat** tab, because they're genuinely different things:

<Cards>
  <Card icon="<Video />" title="Media models">
    Make images, video and audio. You pick one per generation, and it's charged
    per run.
  </Card>

  <Card icon="<Users />" title="Chat models">
    Power the conversation itself — reading, reasoning, writing, using tools.
  </Card>
</Cards>

Both are governed the same way, from the same page, by the same people.

<Cards>
  <Card icon="<Toggle />" title="Who can change this" href="/docs/governance/roles-and-permissions">
    Turning models on and off is a governance permission — admins hold it by
    default.
  </Card>

  <Card icon="<Globe />" title="Where the output goes" href="/docs/concepts/assets-and-visibility">
    Everything a model makes becomes an asset, private to whoever ran it until
    it's shared.
  </Card>
</Cards>