> ## Documentation Index
> Fetch the complete documentation index at: https://www.trybloom.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Generating Images

> The core loop — using your brand, attaching references, generating, and dialing in format.

Once Bloom is connected, generating an image is a conversation: name a brand, describe what you want, and dial in the format from there. This guide walks the core loop with one running example — a launch hero for **Pagecraft**, a SaaS shipping a new AI website builder.

If you haven't connected yet, start with [Getting Started](/docs/mcp/getting-started).

## How it works

```mermaid theme={null}
flowchart LR
  A["Use your brand"] --> B["Attach references"] --> C["Generate"]
```

Every step is something you ask for in plain language; the agent translates it into the right Bloom tools. Once you have an image, [Editing & adapting](/docs/mcp/editing-adapting) covers editing, resizing, and more.

## Using your brand

Every image is scoped to a brand — that's what makes the output on-brand. Name a brand you already have, or onboard a new one from a URL:

```text theme={null}
Use my Pagecraft brand for this.
```

```text theme={null}
Onboard pagecraft.com first, then use it.
```

Every generation draws on the brand's palette, type, logo, and overall aesthetic, so you rarely need to describe those yourself.

## Attaching references

Bloom keeps a searchable library of your brand's images, with semantic search over them — describe a concept and it finds the closest matches. Pulling a few in as references measurably improves results: composition and styling get sharper. Ask for them by concept:

```text theme={null}
Find our past launch graphics in the library to use as reference.
```

By default Bloom shows the matches and lets you pick (handy when you want control). You can also let the agent choose for you:

```text theme={null}
Just pick the two best references yourself and go.
```

If nothing in the library matches, Bloom returns no candidates — that's fine, generation works without references too.

## Generating images

Describe the image. Be specific about subject, mood, and composition; the brand handles the look:

```text theme={null}
Generate three 16:9 variants of a launch hero for our new AI website builder — a clean product mockup with a bold headline and plenty of whitespace. Use the pro model.
```

Generations are asynchronous — usually under a minute or two, depending on the model tier. The agent waits for the result and shows it inline; you don't need to poll or ask "is it done yet?" Ask for up to five variants at once when you're exploring, then keep working with the one you like.

## Specifying format and dimensions

You can ask for any of these controls inline — leave any of them out and Bloom picks a sensible default:

| Control      | Options                                                                 | Notes                                                                         |
| ------------ | ----------------------------------------------------------------------- | ----------------------------------------------------------------------------- |
| Aspect ratio | `1:1`, `16:9`, `9:16`, `4:5`, `3:2`, `2:3`, `3:4`, `4:3`, `5:4`, `21:9` |                                                                               |
| Resolution   | `2K`, `4K`                                                              | `2K` costs 1 credit (default); `4K` costs 2 credits and requires a paid plan. |
| Model        | `fast`, `standard`, `pro`                                               | `pro` is the default and highest quality. See [Choosing Models](/docs/mcp/models). |
| Variants     | 1–5                                                                     | Generate several options at once and pick a favorite.                         |

## Tips

* **Be specific about the subject, not the brand.** Bloom already knows your colors, type, and aesthetic. Spend your words on what's in the frame.
* **Iterate instead of perfecting the first prompt.** Generate, look, then ask for one change at a time.
* **Lean on references.** A couple of good reference images beat a paragraph of description.
* **Generate variants when you're exploring.** It's the fastest way to find a direction.

## Limits & gotchas

* **Content filters.** Some subjects are blocked by the underlying models. If a generation is refused, rephrasing or switching to a more permissive model tier sometimes helps — see [Choosing Models](/docs/mcp/models).
* **Async timing.** Generations usually take under a minute or two depending on the tier. The agent holds the result for you — let it wait rather than asking repeatedly.
* **Typography.** Bloom accurately reproduces your brand's fonts in generated images. Upload your own font files or pick from Google Fonts (natively supported) and they'll be used as-is in your assets. `pro` renders text most faithfully.
* **Credits.** Each image costs 1 credit at `2K` or 2 at `4K`. Variants cost per image, so three variants cost three times one generation. Ask the agent to check your balance any time:

```text theme={null}
How many Bloom credits do I have left?
```