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

# Modalities

> Text, vision, images, video, speech, and embeddings in the Compose inference API.

The catalog describes each model with `input` and `output` modalities. The gateway uses those fields for search, model pages, and operation lists.

Common modalities:

| Modality   | Typical endpoints                                                      |
| ---------- | ---------------------------------------------------------------------- |
| Text       | `/v1/chat/completions`, `/v1/responses`                                |
| Vision     | Text routes with image content parts                                   |
| Images     | `/v1/images/generations`, `/v1/images/edits`, `/v1/responses`          |
| Video      | `/v1/videos/generations`, `/v1/videos/{id}`                            |
| Speech     | `/v1/audio/speech`, `/v1/audio/transcriptions`                         |
| Realtime   | `/v1/responses` with realtime models, `/v1/responses/{id}/input_items` |
| Embeddings | `/v1/embeddings`                                                       |

Specialized outputs such as detections, segmentation, keypoints, and classification appear in catalog metadata for providers such as Roboflow. External configs filter those labels when the target client only accepts a smaller modality enum.

## Browse modalities

```http theme={null}
GET /v1/modalities
GET /v1/modalities/{modality}
GET /v1/modalities/{modality}/operations
GET /v1/modalities/{modality}/operations/{operation}/models
```

Use these endpoints when you need model discovery by task instead of provider name.

## Related

* [Text](/inference/modalities/text)
* [Image](/inference/modalities/image)
* [Video](/inference/modalities/video)
* [Speech](/inference/modalities/speech)
* [Embeddings](/inference/modalities/embeddings)