Regulatory Advisories

By enabling regulatory advisories, AI Scene Analysis screens the asset for imagery that requires a statutory on-screen advisory — such as the tobacco and vaping warnings mandated by the Indian Cigarettes and Other Tobacco Products Amendment Rules, 2023 — and reports each finding against a shot timestamp in the asset description.

Prerequisites

• The feature is off by default and must be enabled for your organization. Contact support before configuring it.
• assetDescription must be enabled, since the findings are written into the asset description file.
• Requires Bitmovin VOD Encoder v2.282 or later.

1. Enabling advisory topics

Set topics to the advisory topics you want the asset screened for. At least one topic is required. Today one topic is supported: TOBACCO, which covers both tobacco and vaping imagery.

aiSceneAnalysis:
  features:
    assetDescription:
      filename: asset-description.json
    regulatoryAdvisories:
      topics:
        - TOBACCO

{
  "aiSceneAnalysis": {
    "features": {
      "assetDescription": {
        "filename": "asset-description.json"
      },
      "regulatoryAdvisories": {
        "topics": ["TOBACCO"]
      }
    }
  }
}

2. Reading the results

AI Scene Analysis reports findings per shot, under shotAdvisories on each shot of a scene. Each entry carries the kind of imagery seen, the model's certainty, and a short explanation.

Example: A shot between 414.2s and 419.8s in which a character smokes:

{
  "scenes": [
    {
      "startInSeconds": 412.5,
      "endInSeconds": 448.0,
      "shots": [
        {
          "startInSeconds": 414.2,
          "endInSeconds": 419.8,
          "shotAdvisories": {
            "status": "ANALYZED",
            "advisories": [
              {
                "category": "TOBACCO",
                "confidence": "HIGH",
                "reason": "A man holds a lit cigarette to his mouth"
              }
            ]
          }
        }
      ]
    }
  ]
}

category distinguishes TOBACCO — smoking, cigarettes, cigars, pipes, tobacco products — from VAPE — e-cigarettes, vape pens and their use. The two are regulated separately in some markets, so they are reported separately.

confidence is HIGH, MEDIUM, LOW or UNKNOWN. Use it to prioritise human review, not as a threshold for discarding advisories: detection is deliberately tuned to flag an uncertain cue rather than miss it, so advisories below HIGH are expected.

3. Shots that carry no verdict

When regulatory advisories are enabled, every shot in the asset carries a shotAdvisories object. status tells you whether a verdict was actually reached.

• ANALYZED with an empty advisories array — AI Scene Analysis assessed the shot and found nothing.
• BLOCKED — the model's safety filter rejected the request, so no verdict could be produced. The filter reacts to the frames themselves, not to the advisory question — a cigarette lit during a violent or sexual scene can be rejected on that surrounding content. AI Scene Analysis reports the shot conservatively, with a TOBACCO advisory at UNKNOWN confidence. That advisory is an assumption made in the absence of a verdict, not an observation. Review these shots manually.

{
  "startInSeconds": 512.0,
  "endInSeconds": 517.4,
  "shotAdvisories": {
    "status": "BLOCKED",
    "advisories": [
      {
        "category": "TOBACCO",
        "confidence": "UNKNOWN",
        "reason": "Model refused to analyze this shot; treated as a positive advisory."
      }
    ]
  }
}

A clean asset is therefore one where every shot reports ANALYZED with no advisories. BLOCKED shots are the only gap in coverage, and they are surfaced rather than passed over as clean.