---
title: "Image generation"
description: "Generate images with Laravel AI's Image API and the models Cloudflare hosts, with reference images on FLUX.2."
---

> Documentation Index
> Fetch the complete documentation index at: https://laravel-cf-workersai-docs.mdick85.workers.dev/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Image generation

The package adds Workers AI to Laravel AI's `Image` API. It needs no extra config, and works on the [direct route and the gateway route](/docs/concepts/routes/).

To send an image to a chat model, see [Image input](/docs/guides/image-input/) instead.

## Generate an image

```php
use Laravel\Ai\Image;

$image = Image::of('A green apple on a wooden table.')
    ->landscape()
    ->quality('medium')
    ->generate(provider: 'workers-ai');

$path = $image->store('images', 'public');

$bytes = (string) $image;
$mime = $image->images->first()->mime;
```

## Size and quality

| Method | Sends |
| --- | --- |
| `square()` | 1024 x 1024 |
| `portrait()` | 832 x 1248 |
| `landscape()` | 1248 x 832 |
| `size('512x384')` | The width and height you give |
| `quality('low')` | 4 inference steps |
| `quality('medium')` | 8 inference steps |
| `quality('high')` | 20 inference steps |

With no size or quality, Cloudflare uses the model's defaults. More steps take longer and cost more. `flux-1-schnell` allows at most 8 steps, so the package caps it there.

## Choose a model

The default is `@cf/black-forest-labs/flux-2-klein-4b`. To change it, set `models.image.default` on the provider entry in `config/ai.php`, or pass `model:` to `generate()`. Each model has its own limits. See the [Cloudflare model catalog](https://developers.cloudflare.com/workers-ai/models/).

| Model | Size | Reference images |
| --- | --- | --- |
| `@cf/black-forest-labs/flux-2-klein-4b`, `flux-2-klein-9b`, `flux-2-dev` | Yes | Yes |
| `@cf/black-forest-labs/flux-1-schnell` | No. A size throws. | No |
| `@cf/bytedance/stable-diffusion-xl-lightning`, `@cf/stabilityai/stable-diffusion-xl-base-1.0`, `@cf/lykon/dreamshaper-8-lcm` | Yes | No |
| `@cf/leonardo/lucid-origin`, `@cf/leonardo/phoenix-1.0` | Yes | No |

To use a model's other inputs, add them with `withProviderOptions()`. It needs `laravel/ai` 1.0 or later.

```php
Image::of('A green apple.')
    ->withProviderOptions(['seed' => 42])
    ->generate(provider: 'workers-ai');
```

## Use reference images

FLUX.2 models take up to 4 reference images. Cloudflare says each one must be smaller than 512 x 512. In the prompt, refer to them by position or in plain words.

```php
use Laravel\Ai\Files\Image as ImageFile;

$image = Image::of('Take the subject of image 1 and style it like image 0.')
    ->attachments([
        ImageFile::fromPath(storage_path('app/style.png')),
        ImageFile::fromPath(storage_path('app/subject.png')),
    ])
    ->generate(provider: 'workers-ai');
```

A fifth image throws before a request. So does an attachment on a model that does not take one.

## Omit the provider name

Laravel AI's `default_for_images` is `gemini` until you change it. Set it in `config/ai.php` to omit `provider:`:

```php
'default_for_images' => 'workers-ai',
```

## FLUX.2 and AI Gateway

FLUX.2 models send multipart, so with a gateway they use the gateway's own URL. On an authenticated gateway, they need `gateway_token`, on both the gateway route and the header route. Other image models need only `key`. See [Non-JSON models](/docs/concepts/routes/#non-json-models).

## Errors

Cloudflare's content filter can reject a prompt with `400 Input prompt contains NSFW content`, and it can flag harmless prompts. In tests on 2026-11-05, it rejected "A green apple on a wooden table." on `phoenix-1.0` in 3 of 8 runs. The same prompt passed on the other models tested. Try again, reword the prompt, or use another model.

## Test your code

```php
use Laravel\Ai\Image;

Image::fake();

// run your code, then:
Image::assertGenerated(fn ($prompt) => $prompt->contains('apple'));
```

## Cost

Cloudflare bills images in neurons. One `flux-1-schnell` image at 1 step cost 57.6 neurons in a test on 2026-11-05. The package reports the figure with `WorkersAiUsageReported`. See [Cost tracking](/docs/guides/cost-tracking/).

Source: https://laravel-cf-workersai-docs.mdick85.workers.dev/docs/guides/images/index.mdx