Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/phpaisdk/ollama/llms.txt

Use this file to discover all available pages before exploring further.

OllamaImageModel implements ImageModelInterface and sends requests to Ollama’s experimental OpenAI-compatible /images/generations endpoint. Because Ollama’s image generation support is still evolving, the adapter deliberately restricts the outgoing request to only the fields Ollama currently accepts — all other fields are stripped before the request is sent.

​
Namespace

use AiSdk\Ollama\Models\OllamaImageModel;
Image generation is experimental in Ollama. The available models, supported sizes, and API stability all depend on the Ollama server version you are running. Behaviour may change without notice across Ollama releases.
Generate::speech() is intentionally not implemented in this provider. Ollama does not expose an OpenAI-compatible speech endpoint, so no OllamaSpeechModel exists.
This class is not instantiated directly. Use Ollama::model() to obtain a model instance and the Generate facade to dispatch requests.

​
URL

Requests are sent to: 
{baseUrl}/images/generations
The baseUrl defaults to http://localhost:11434/v1.

​
Method

​
generate(ImageRequest $request): ImageResponse

Builds a restricted request body and POSTs it to /images/generations, then parses the response with ImageResponseParser.

​
Allowed request fields

OllamaImageModel uses ImageRequestBuilder to assemble a full request body, then filters it down to only the four keys Ollama accepts:
​
model
string
required
The model identifier string — for example 'x/z-image-turbo'.
​
prompt
string
required
The text prompt describing the image to generate.
​
size
string
The output image dimensions as a 'WxH' string — for example '1024x1024'. If not set by the caller, the adapter defaults this to '1024x1024' automatically.
​
response_format
string
Always set to 'b64_json' by the request builder. The adapter only handles base64-encoded image responses; URLs are not requested.

​
Fields that are NOT forwarded

Even if you set these values in your Generate call, they are stripped before the request reaches Ollama:
SDK callReason omitted
->count(n)n / count is not forwarded (includeCount: false)
->seed(n)seed is not forwarded (seedParameter: null)
->quality('hd')Not in Ollama’s allowed field list
->style('vivid')Not in Ollama’s allowed field list
Provider optionsNot merged (includeProviderOptions: false)

​
Response

generate() returns an ImageResponse parsed by ImageResponseParser.
​
output
ImageOutput
The first generated image wrapped in an ImageOutput object.
​
output.bytes()
string
The raw image bytes, base64-decoded from the b64_json field returned by Ollama. Write directly to a file or pass to an image-processing library.

​
Error Conditions

generate() raises an InvalidResponseException in the following cases: Missing or non-array data field — if $payload['data'] is absent or is not an array, the adapter throws before delegating to the response parser: 
Provider [ollama] returned no generated images.
Empty data array — if $payload['data'] is an empty array, ImageResponseParser raises an InvalidResponseException during parsing. Both conditions can occur when targeting a model that does not support image generation, or when the Ollama server version does not yet implement the endpoint.

​
Usage Example

use AiSdk\Generate;
use AiSdk\Ollama;

Ollama::create();

$result = Generate::image('A small robot on a desk')
    ->model(Ollama::model('your-image-model'))
    ->size('1024x1024')
    ->run();

$bytes = $result->output->bytes();
file_put_contents('robot.png', $bytes);

​
Default size behaviour

If you omit ->size(), the adapter automatically sets size to '1024x1024' in the outgoing request: 
Ollama::create();

// size defaults to '1024x1024' — no need to set it explicitly
$result = Generate::image('A mountain at sunrise')
    ->model(Ollama::model('your-image-model'))
    ->run();

file_put_contents('mountain.png', $result->output->bytes());

​
Fields stripped by the adapter

The following example demonstrates that count, seed, and provider options are silently ignored — only the four allowed fields reach Ollama: 
Ollama::create();

// count(2), seed(42), and the raw provider options below are all stripped.
// The body sent to Ollama contains only: model, prompt, size, response_format.
$result = Generate::image('A tiny robot')
    ->model(Ollama::model('x/z-image-turbo'))
    ->count(2)
    ->size('1024x1024')
    ->seed(42)
    ->providerOptions('ollama', [
        'raw' => [
            'quality' => 'high',
            'style'   => 'vivid',
            'user'    => 'user_123',
        ],
    ])
    ->run();

file_put_contents('robot.png', $result->output->bytes());

Build docs developers (and LLMs) love

Get started for freeTalk to us