# Image

> Generate an image in Endless: pick a model, write a prompt, add reference pictures, set the model's own controls and see the cost before you run it.

Source: https://endless.okinawa/docs/create/image

**Image** is a composer with a feed of your results above it. You set up one run (model, prompt,
reference pictures, settings) and press **"Generate"**. The composer keeps your settings, so for
the next run you change only what should differ.

The **Manual** tab is this composer. The **Agent** tab sends your prompt to a new [chat](/docs/chat)
with `@image` in it, where Endless picks the model.

<ImageZoom src="https://static.endless.software/docs/screenshots/image-studio-1ac57775.webp" alt="The Image page: Manual and Agent tabs and a Show library button in the header, earlier results in the feed, and the docked composer with Generate image, the prompt field, a 16:9 Output chip, the advanced parameters control, the selected model, the credit estimate and the Generate button" width="1280" height="720" />

## Generate an image

<Steps>
  <Step>
    ### Write a prompt

    Type in &#x2A;"Describe the image you want to create…"*, or pick one of the starting ideas to load
    its prompt.
  </Step>

  <Step>
    ### Choose a model

    The model control is in the composer's footer. The list is grouped by provider, has a search
    box, and shows each model's price. The model decides every other control on the page. See
    [Models](/docs/concepts/models).

    <ImageZoom src="https://static.endless.software/docs/screenshots/image-model-picker-9991d799.webp" alt="The model picker open above the Image composer: a provider rail on the left, a search field, and models grouped by provider (GPT Image 2.5 checked, FLUX 3 Image, Ideogram 4.5), each row with its price in credits and what it is good at" width="720" height="465" />
  </Step>

  <Step>
    ### Add references (optional)

    **"Add reference"** opens your assets. You can also paste or drop a picture into the prompt
    field, or drag an earlier result from the feed onto the composer.
  </Step>

  <Step>
    ### Set the output and parameters

    **"Output"** sets shape and resolution (the chip reads e.g. **16:9 · 1080p*&#x2A;). &#x2A;*"Advanced
    parameters"*&#x2A; shows the model's other settings; &#x2A;*"Reset all"** restores its defaults.
  </Step>

  <Step>
    ### Check the estimate and generate

    The number next to &#x2A;*"Generate"** is the expected cost. A placeholder appears in the feed at
    once.
  </Step>
</Steps>

<Callout icon="<WandSparkle />">
  **"Enhance prompt"*&#x2A; rewrites your prompt with more detail, using your
  reference images too. It costs no credits. &#x2A;*"Restore original"** brings your
  text back.
</Callout>

## Controls depend on the model

Each model accepts different settings, and the composer shows only those. Switching model adds,
removes or changes controls.

| What you see                                                                  | Meaning                                                                  |
| ----------------------------------------------------------------------------- | ------------------------------------------------------------------------ |
| *"This model doesn't take reference assets"&#x2A; on &#x2A;*"Add reference"** | The model accepts no references                                          |
| *"This model's references are already filled"*                                | All reference slots are used                                             |
| A slot marked &#x2A;*"Required"** (e.g. a base image)                         | **"Generate"** won't run until it's filled                               |
| *"This model only uses the references, no prompt needed"*                     | The prompt field is disabled                                             |
| No &#x2A;*"Advanced parameters"** toggle                                      | The model has no extra settings                                          |
| *"2 files didn't fit this model's inputs"*                                    | You picked more references than the model takes; the extras were dropped |

## When a model isn't available

* **Locked to a plan:*&#x2A; shown dimmed with a lock and &#x2A;"Only available from the … plan"*.
  Choosing it opens billing; nothing runs or is charged.
* **Turned off by the workspace:** missing here, in chat and in `@&#x60; mentions. With every model
  off, the control reads &#x2A;"No models available. Ask an admin to enable one in workspace
  settings."*
* **Removed from the catalog:*&#x2A; earlier results stay in your library, but &#x2A;*"Regenerate"*&#x2A; says
  &#x2A;"This model is no longer available"*.

## Cost

The estimate uses the same rules as the charge, but it prices the outputs you asked for, and the
charge prices the outputs that come back. Models without a fixed price show &#x2A;*"from"** a number
in the list; the composer shows the real figure once your inputs are set. A failed run is not
charged. If the workspace or your own limit can't cover a run, it is refused before it starts.
See [Credits and spending](/docs/concepts/credits-and-spending).

## Reuse a result

* **"Show library"*&#x2A; opens your library beside the page. &#x2A;*"Use as reference"*&#x2A; (the &#x2A;*+** on a
  picture) or dragging a result onto the composer adds it as a reference.
* Open a result to see its prompt and model. &#x2A;*"Regenerate"*&#x2A; runs the same setup again;
  &#x2A;*"Reuse parameters"** (in its menu) loads the setup into the composer so you can change it
  first.
* The feed is a view of your [library](/docs/create/where-output-lands), not a separate store.

## Cancelling a run

Open a running placeholder and choose &#x2A;*"Cancel generation"**. If the provider hadn't started,
nothing is charged. If it had, the run may finish anyway: it is charged and the result goes to
the trash instead of the feed.