This guide walks you through the process of storing and retrieving embeddings using the Mixpeek API. Follow the steps below to get the full payload on a per feature basis, along with the corresponding named vectors.

​
Step 1: Index

First, index your video using the Mixpeek API. You can do this by sending a request to the Index Video URL endpoint.

import requests
import json

url = "https://api.mixpeek.com/index/videos/url"

payload = json.dumps({
  "url": "video.mp4",
  "collection_id": "example"
})
headers = {
  'Authorization': 'Bearer API_KEY',
  'Content-Type': 'application/json'
}

response = requests.request("POST", url, headers=headers, data=payload)

You’ll then get a task_id in the response. Use this ID to poll the status of the task.

{
  "message": "Video URL processing queued",
  "task_id": "857056f8-1f0d-4cad-aa37-96757e12872e"
}

​
Step 2: Poll Status

After indexing, poll the status of the task to ensure it has completed. Use the Get Task endpoint.

url = "https://api.mixpeek.com/tasks/3f0189c2-2f49-4119-851d-56a5701436a9"

response = requests.request("GET", url, headers=headers)

Which should reeturn this response:

{
  "asset_id": "60f391eb-b5ea-41cd-ba75-98d1bca25432",
  "task_id": "3f0189c2-2f49-4119-851d-56a5701436a9",
  "status": "DONE"
}

​
Step 3: Get Asset with Features

Once the task is complete, retrieve the asset with its features using the Get Asset with Features endpoint.

url = "https://api.mixpeek.com/assets/60f391eb-b5ea-41cd-ba75-98d1bca25432/features"

response = requests.request("GET", url, headers=headers, data=payload)

{
  "asset": {
    "asset_id": "60f391eb-b5ea-41cd-ba75-98d1bca25432",
    "collection_id": "example",
    "status": "DONE",
    "url": null,
    "metadata": {
      "asset": {},
      "author": "user",
      "preview_url": null
    },
    "error": null,
    "modality": "text",
    "task_id": "3f0189c2-2f49-4119-851d-56a5701436a9",
    "unique_hash": null,
    "updated_at": "2024-11-07T14:31:56.428000",
    "created_at": "2024-11-07T14:31:55.473000",
    "score": null
  },
  "features": {
    "video": [],
    "image": [],
    "text": [
      {
        "feature_id": "98b693c0-6fc1-4420-97bd-6c4506356230",
        "modality": "text",
        "asset_id": "60f391eb-b5ea-41cd-ba75-98d1bca25432",
        "collection_id": "example",
        "metadata": {
          "author": "user"
        },
        "text": "two people inside a car",
        "created_at": "2024-11-07T14:31:56.053031Z"
      }
    ],
    "audio": []
  }
}

Iterate through each modality’s list of features to access detailed information.

​
Step 4: Get Features with Vector

For each feature, retrieve the vectors using the Get Feature endpoint.

url = "https://api.mixpeek.com/features/98b693c0-6fc1-4420-97bd-6c4506356230?include_vectors=true"


response = requests.request("GET", url, headers=headers)

​
Example Payload

The following is an example of the full payload you can expect to receive:

{
  "created_at": "2024-11-06T18:19:43.340820Z",
  "asset_id": "cfa6c2a4-b695-4fc2-84df-868391b218e0",
  "metadata": {},
  "feature_id": "3086f1ee-62f1-49cf-b859-b553044aebed",
  "modality": "image",
  "vectors": {
    "multimodal": [-0.022097813, 0.021723306]
  }
}

This payload provides the full details of the asset and its features, including the named vectors for each modality.

You can then store these vectors in your database for future use, see example database integrations: Database Integrations

Was this page helpful?

RerankingFine-Tuning