LLM
Send images to LLMs (vision) via API
In POST /llm/chat/completions, a message's content can mix text and image_url parts, as in OpenAI. Models with supports_vision in GET /llm/models accept images, and Connect translates image_url for Anthropic, so the code is the same.
The request
POST /api/llm/chat/completions
Authorization: Bearer <merchant token>
{
"model": "anthropic/claude-sonnet-5-5",
"max_tokens": 512,
"messages": [
{
"role": "user",
"content": [
{
"type": "text",
"text": "Describe this ad creative."
},
{
"type": "image_url",
"image_url": {
"url": "https://example.com/creative.jpg"
}
}
]
}
]
}What you can do
- Ask the model to describe or analyze an image.
- Mix text and several images in the same message.
- Check in GET /llm/models which models have supports_vision.
- Combine images with stream and fallback.
Rules and limits
- image_url parts are only accepted if every model in the call (model and each item of models) has supports_vision. Otherwise: 400 llm_unsupported_field.
- On Anthropic, audio or file parts return 400 llm_unsupported_field; images through image_url are translated.
- The request body is at most 1 MB (413 llm_request_too_large).
- max_tokens is still required and cannot exceed the account cap or the models' max_output_tokens.
- Images count as input tokens, billed by the provider on your key.
Main fields
| Field | Type | Required | Description |
|---|---|---|---|
| model | string | Yes | Model id from GET /llm/models, in provider/model form. |
| messages | array | Yes | The conversation. content can be text or text and image_url parts. |
| max_tokens | integer | Yes | Required (or max_completion_tokens). Up to the account cap and each model's max_output_tokens in the call. |
| models | array | No | Fallback: up to 3 models from GET /llm/models, tried in order if model fails before the first token. |
Who it is for
Ad creative analysis
Your product sends a campaign's creative and gets an assessment of the visual and the copy.
Alt text and captions
Generate image descriptions for customers' posts straight from your SaaS.
Agents that read screenshots and charts
An agent receives a dashboard screenshot and explains the numbers in text.
Frequently asked questions
Which models accept images?
Those with supports_vision true in GET /llm/models. Sending image_url to a model without it returns 400 llm_unsupported_field.
How do I send the image?
As in OpenAI: a content part with type image_url and the image_url object. Connect translates it for Anthropic.
Is there a size limit?
The whole request body can be up to 1 MB. Above that the response is 413 llm_request_too_large.
See Connect applied to your product.
Try it for 15 days with no card, or walk our team through your use case.
