Configuring each feature

Following are details about how to configure each feature (output) that you include in a Elemental Inference feed.

Configuring event clipping

In Callback config, you can enter a string that you want Elemental Inference to always include in the event clipping metadata for this output. This information is useful when you later work with Elemental Inference events in Amazon EventBridge. You will be able to filter events using this information, in order to find the events for one feed. The string might identify the sports event in the feed, for example.

Aligning clips to plays with Sports Data

To use Sports Data, you find the game in the sports data feed and attach it to an event clipping output. Elemental Inference then aligns the clips for that output to the plays in the game.

Supported sports and leagues

Sports Data supports the following sports and leagues.

For each sport, Elemental Inference aligns clips to the following events. The tag is the value that identifies the event in the game data.

Basketball events

American football events

Step 1: Find your game

Use the SearchFixtures operation to find the game that you want to align clips to. You provide a sport and a date range, and optionally filter by team. Each game in the response includes a fixtureId that you attach to your event clipping output in the next step.

The following example searches for basketball games on 2026-03-03 and 2026-03-04 that involve the Comets.

$ awscurl --service "elemental-inference" --region <region> \
  -X POST "https://<data-endpoint>/v1/fixtures" \
  -d '{
    "sport": "basketball",
    "startDate": "2026-03-03",
    "endDate": "2026-03-04",
    "filters": [{ "name": "COMPETITOR", "values": ["Comets"] }]
  }'

The response returns the matching games. Note the fixtureId of the game that you want to clip.

{
  "fixtures": [
    {
      "fixtureId": "NwrW5B17Rw9W25g1jPlobSGdQfBOKyD3227bcsRK0w3Bq7aG2xRrO8qx2XEjl6R1sOH2kwyjbojN4Rk7qHiF2w",
      "name": "Northport Comets vs. Southport Meteors",
      "fixtureGroup": "Regular Season",
      "scheduledStart": "2026-03-04T03:00:00Z",
      "status": "Scheduled",
      "competitors": [
        { "name": "Meteors", "isHome": true },
        { "name": "Comets", "isHome": false }
      ]
    }
  ]
}

Request parameters

Game status

The status of a game tells you whether it is safe to attach.

You can search for past games the same way. Set startDate and endDate to dates in the past, and attach a Completed game to clip a game that has already aired.

Filtering by team

When you filter by team, every word that you send must appear in the game's team names. Matching is case-insensitive and matches partial words, but does not do fuzzy matching. You can search by city name even though the response returns only the team name.

For the game Northport Comets vs. Southport Meteors, the values comets, Comet, and Northport all match, but Cometz does not.

Words match independently, so Northport Meteors also matches the game in the previous example.

Step 2: Attach the game to your output

Attach the game to an event clipping output by including the fixtureId in the output's dataSourceConfiguration. You can do this when you create the feed, or on an existing feed by using UpdateFeed. The fixtureId is the only value you provide, and you can attach one game to each event clipping output.

The following example attaches a game to an event clipping output when creating a feed.

aws elemental-inference create-feed \
  --name "my-feed" \
  --outputs '[{
    "name": "clipping",
    "status": "ENABLED",
    "outputConfig": {
      "clipping": {
        "dataSourceConfiguration": {
          "fixtureId": "NwrW5B17Rw9W25g1jPlobSGdQfBOKyD3227bcsRK0w3..."
        }
      }
    }
  }]'

Checking that it works

Clips appear only after both of the following are true: the game has started, and your channel is sending media. You can confirm the results as follows:

If you don't see clips yet, check the following.

Configuring smart crop

Smart crop has no required configuration. Optionally, you can add graphic composition to a smart crop output to detect known graphics, such as scoreboards and advertisements, in your source media. You provide one or more reference images, called templates, and Elemental Inference reports, for each analyzed frame, whether each graphic is present and where it appears as a bounding box. Elemental Inference returns graphic composition results as part of the smart crop metadata. For more information, see Metadata for graphic composition.

You can configure one to four template groups of reference images, where each group represents a single graphic to detect. Configuring at least one group enables graphic composition in the output. Each group has the following settings:

Template image recommendations

For the best detection performance, follow these recommendations when you prepare your template images:

CLI example

The following example shows how to include a smart crop output that detects two graphics, a scoreboard and ads, when creating a feed using the CLI:

aws elemental-inference create-feed \
  --name "my-feed" \
  --access-role-arn "arn:aws:iam::111122223333:role/my-ei-access-role" \
  --outputs '[{
    "name": "crop",
    "status": "ENABLED",
    "outputConfig": {
      "cropping": {
        "templateGroups": [
          {
            "name": "scoreboard",
            "templateUris": [
              "s3://amzn-s3-demo-bucket/scoreboard-v1.png",
              "s3://amzn-s3-demo-bucket/scoreboard-v2.png"
            ]
          },
          {
            "name": "ads",
            "templateUris": [
              "s3://amzn-s3-demo-bucket/ads.png"
            ]
          }
        ]
      }
    }
  }]'

Configuring smart subtitles

Smart subtitles uses automatic speech recognition (ASR) to generate TTML subtitles from the audio in your source media. Configure the following settings for the smart subtitles output:

CLI example

The following example shows how to include a smart subtitles output when creating a feed using the CLI:

aws elemental-inference create-feed \
  --name "my-feed" \
  --outputs '[{
    "name": "subtitles",
    "status": "ENABLED",
    "outputConfig": {
      "subtitling": {
        "language": "eng",
        "aspectRatio": {"width": 16, "height": 9},
        "profanityFilter": "DISABLED"
      }
    }
  }]'

Managing dictionaries

A dictionary contains custom words and phrases that the ASR engine might not recognize, such as brand names, technical terms, or proper nouns. You can reference a dictionary when configuring a smart subtitles output to improve transcription accuracy for domain-specific terminology.

Use the Elemental Inference dictionary API operations to manage dictionaries:

Dictionary validation rules

When creating or updating a dictionary, the following validation rules apply: