Get Detection Result | Resemble

GET https://app.resemble.ai/api/v2/detect/{uuid}

Retrieve the deepfake detection result using the provided UUID. Returns the final metrics for the detection job. Media-type-specific keys appear depending on the input.

curl --request GET 'https://app.resemble.ai/api/v2/detect/DETECT_UUID' \
  -H 'Authorization: Bearer YOUR_API_TOKEN'

Query Parameters

Field	Type	Default	Description
`experts`	`boolean`	`false`	When `true`, return every successfully completed Intelligence result in an array. When omitted or `false`, return one completed result when available; otherwise, return the current processing or failed result.

item.status reports the requested Detect work. When detect_watermark=true was used, an otherwise successful Detect remains processing until the nested Watermark analysis is terminal. A core detection failure still reports failed. Detect and Intelligence run separately, so item.status: "completed" does not mean Intelligence has finished.

When experts is omitted or false, inspect item.intelligence.status and item.intelligence.description for the returned Intelligence result. It can still be processing with a null description after Detect completes, and its UUID can change as additional Intelligence results finish. With experts=true, the array contains only successfully completed results available at the time of the request; an empty array means no result has completed successfully yet, not that Intelligence was not requested.

Response Details

status - Status of the requested Detect work (processing, completed, or failed). An otherwise successful Detect waits for requested Watermark analysis; a core detection failure still reports failed. The status remains independent of Intelligence processing.
metrics - Always includes a label (fake/real), score array with predictions per chunk, consistency field (for audio/video), and aggregated_score
image_metrics - Present for still images, contains type, label, score, visualization image (if requested), and nested analysis children
video_metrics - Present for videos, includes nested detections across frames with timestamps, certainty scores, and hierarchical results
modality - For video detects, which components were requested: audio, video, or all (default). When a single modality was requested, the skipped component’s metrics are absent (video_metrics for audio, metrics for video) and the detect still reaches completed
face_only - Effective face-only setting. It is true only for video detects that included visual analysis; audio, image, and audio-only video detects return false
audio_source_tracing - Present when audio is labeled as fake. Contains label identifying the likely source (e.g., resemble_ai, elevenlabs) or real if audio is not synthetic, and error_message (null on success)
intelligence - Present when intelligence analysis was requested. By default, contains one Intelligence object with uuid, status (processing, completed, or failed), error_message, description, created_at, and detect_uuid. The API returns a completed result when one is available; if none have completed successfully, it returns the current processing or failed result. With experts=true, this field is an array containing all successfully completed results and excludes processing or failed results.
watermark - Present only when detect_watermark=true was requested. Contains status plus completed Resemble detector metrics and the optional synthid boolean. A failed child includes error_message and does not change the deepfake verdict.
c2pa_manifest - Present for audio, image, and video detects after Content Credentials verification completes. validation_state can be Valid, NotPresent, or Unavailable; Unavailable means C2PA verification could not complete, but the detect result is still valid.
zero_retention_mode - Boolean indicating whether Zero Retention Mode was enabled for this detect
file_deleted_at - ISO 8601 timestamp of when the file was purged, or null if not yet deleted. Use this to determine whether the source media has been purged from storage.
audio_url - Alias of url. Both are null when Zero Retention Mode is enabled

Retrieve All Successful Intelligence Results

curl --request GET 'https://app.resemble.ai/api/v2/detect/DETECT_UUID?experts=true' \
  -H 'Authorization: Bearer YOUR_API_TOKEN'

With experts=true, item.intelligence contains every successfully completed Intelligence result available at the time of the request. The array is empty when no result has completed successfully:

{
  "success": true,
  "item": {
    "uuid": "DETECTION_UUID",
    "status": "completed",
    "intelligence": [
      {
        "uuid": "INTELLIGENCE_UUID_A",
        "status": "completed",
        "error_message": null,
        "description": {
          "speaker_info": "Adult speaker",
          "language": "English"
        },
        "created_at": "2024-01-15T10:30:05Z",
        "detect_uuid": "DETECTION_UUID"
      },
      {
        "uuid": "INTELLIGENCE_UUID_B",
        "status": "completed",
        "error_message": null,
        "description": {
          "speaker_info": "Adult speaker",
          "language": "English"
        },
        "created_at": "2024-01-15T10:30:06Z",
        "detect_uuid": "DETECTION_UUID"
      }
    ]
  }
}

When experts is omitted or false, the singular Intelligence UUID may change across polls as additional results finish. Detect callbacks also contain a single Intelligence result and allow Intelligence processing to continue up to the callback timeout before choosing what to include. The callback includes a completed result when one is available, or the current processing or failed result when none have completed successfully. A callback is not resent if another Intelligence result completes later.

{
  "success": true,
  "item": {
    "uuid": "DETECTION_UUID",
    "media_type": "audio",
    "status": "completed",
    "modality": "all",
    "face_only": false,
    "metrics": {
      "label": "fake",
      "score": ["0.9", "0.8", "0.7"],
      "consistency": "0.85",
      "aggregated_score": "0.566"
    },
    "image_metrics": {
      "ifl": {
        "score": 1,
        "heatmap": "https://..."
      },
      "type": "FinalResult",
      "label": "Fake",
      "image": "https://...",
      "score": 1,
      "children": []
    },
    "video_metrics": {
      "label": "Fake",
      "score": "0.9",
      "certainty": "0.9",
      "treeview": "...",
      "children": [
        {
          "type": "VideoResult",
          "conclusion": "Fake",
          "score": "0.99",
          "certainty": "0.9",
          "children": [
            {
              "type": "ImageResult",
              "conclusion": "Fake",
              "score": "0.9",
              "certainty": "0.95",
              "timestamp": "0.23",
              "children": [
                {
                  "type": "Segment",
                  "conclusion": "Fake",
                  "score": "0.9",
                  "certainty": "0.995"
                }
              ]
            }
          ]
        }
      ]
    },
    "audio_source_tracing": {
      "label": "resemble_ai",
      "error_message": null
    },
    "intelligence": {
      "uuid": "INTELLIGENCE_UUID",
      "description": "Adult male speaker with confident tone discussing...",
      "created_at": "2024-01-15T10:30:05Z",
      "detect_uuid": "DETECTION_UUID"
    },
    "c2pa_manifest": {
      "validation_state": "NotPresent"
    },
    "watermark": {
      "status": "completed",
      "metrics": {
        "overall_status": "absent",
        "detected_model_versions": [],
        "coverage_complete": true,
        "has_watermark": {
          "channel_0": false
        },
        "confidence": 0.0,
        "model_results": [
          {
            "model_version": "perth_v1",
            "status": "succeeded",
            "detected": false,
            "has_watermark": {
              "channel_0": false
            },
            "confidence": 0.0
          },
          {
            "model_version": "perth_v2",
            "status": "succeeded",
            "detected": false,
            "has_watermark": {
              "channel_0": false
            },
            "confidence": 0.0
          }
        ],
        "synthid": false
      }
    },
    "url": "https://storage.example.com/audio/suspicious.wav",
    "audio_url": "https://storage.example.com/audio/suspicious.wav",
    "filename": "suspicious.wav",
    "duration": 8.2,
    "zero_retention_mode": false,
    "file_deleted_at": null,
    "created_at": "2024-01-15T10:30:00Z",
    "updated_at": "2024-01-15T10:30:05Z"
  }
}

metrics always reflects the aggregate prediction and confidence per chunk.
image_metrics appears for still images and contains visualization assets if requested.
video_metrics nests detections across frames with timestamps and certainty percentages.
audio_source_tracing is only returned when audio is labeled as fake. Use the standalone Audio Source Tracing endpoints to access all source tracing reports.
watermark is omitted entirely for legacy requests where watermark analysis was omitted or false. If present, metrics.synthid is omitted when SynthID was unavailable; omission must not be interpreted as a negative verdict.

Zero Retention Mode Response

When zero_retention_mode was enabled on the original request, media URLs are redacted and the filename is tokenized:

{
  "success": true,
  "item": {
    "uuid": "DETECTION_UUID",
    "media_type": "audio",
    "status": "completed",
    "face_only": false,
    "metrics": {
      "label": "fake",
      "score": ["0.9", "0.8", "0.7"],
      "consistency": "0.85",
      "aggregated_score": "0.566"
    },
    "url": null,
    "audio_url": null,
    "filename": "redacted_abc123def4567890abcd1234.wav",
    "duration": 8.2,
    "c2pa_manifest": {
      "validation_state": "NotPresent"
    },
    "zero_retention_mode": true,
    "file_deleted_at": "2024-01-15T10:31:00Z",
    "created_at": "2024-01-15T10:30:00Z",
    "updated_at": "2024-01-15T10:30:05Z"
  }
}

Error Response (404)

{
  "success": false,
  "message": "Detect not found"
}