IP Detection - Screen Generations for Intellectual Property Risk

IP Detection is an Enterprise add-on that checks every generation before it runs and blocks the ones that carry intellectual-property risk: a named copyrighted character, a trademarked logo, the likeness of a real public figure, or the signature style of a named artist.

It is configured once for your whole organization and applies to generations from the web application and from the API alike. This is different from the content policies enforced by the model providers themselves, which Scenario does not control (see Content Policy & Copyright). IP Detection is your own policy, on your own terms.

1. What IP Detection Does

When a generation is submitted, IP Detection reads the prompt and the input images and runs them through the filters your organization has enabled. Each filter returns a verdict, with the entities it recognized, a short reason, and a confidence score.

What happens next depends on your enforcement setting: the job is either allowed through, or blocked before any generation cost is incurred.

2. Requirements and Permissions

• Available on the Enterprise plan.

• Only an organization admin can change the settings. Other members see the page in read-only form.

• Settings live in Organization > Add-ons > IP Detection. Filters are organization-wide, so there is one policy to maintain rather than one per project.

3. Turn On Enforcement

Open Organization > Add-ons > IP Detection, expand Enforcement, and choose how strict you want to be.

A blocked generation fails immediately with a message explaining that the request was blocked for potential intellectual-property risk. Nothing is generated and no generation cost is charged.

4. Choose Your Filters

Expand Filters and turn on the built-in categories you care about. All four start switched off, so turning enforcement on without enabling at least one filter screens nothing: the Filters section shows Missing as a reminder.

5. Create a Custom Filter

Built-in categories cover the common cases. For the IP that is specific to your studio, your licensors or your client, add your own: up to six custom filters per organization.

1. In Filters > Custom filters, select Create filter.

2. Give it a name your team will recognize, for example "Disney characters" or "Client X mascots".

3. Write the instruction: a plain-language description of what to look for, up to 2,000 characters. Be specific about what should be flagged, and about what should not.

4. Optionally select Draft from reference images and pick up to eight images from your library, or upload them. Scenario writes a first draft of the instruction from what it sees. Reference images are used only to write that draft, so review and edit the text before saving.

5. Select Create filter. New filters are active as soon as enforcement is on, and you can toggle, edit or delete each one later.

Editing an instruction takes effect on the next generation, so you can tighten a filter that is too broad without waiting.

6. Opt a Single Project Out

Some work does not need screening: an internal experiment, a licensed pipeline where the IP is yours to begin with. Open the project, go to Add-ons > IP Detection, and turn on Disable IP Detection. Generations in that project are no longer screened and no longer charged the detection fee.

Two things to keep in mind:

• This is an opt-out only. A project cannot pick its own strictness, and it cannot enable screening that the organization has not turned on.

• Only an organization admin can set it, and filters remain organization-wide.

7. What It Costs

Screening is billed in Compute Units, separately from the generation itself:

• 1 CU per generation that IP Detection screens.

• 1 CU for each input image it analyzes. A prompt-only generation costs 1 CU to screen; the same generation with two reference images costs 3 CU.

The fee is charged whenever the check runs, including when the result is a block, and it is reported as its own IP Detection line in your usage rather than folded into the generation cost. Inputs that are not images, and images that cannot be read, are not charged.

8. Run a Check From the API

If your organization has enforcement on, every API generation is screened automatically. You can also ask for a check explicitly, which is useful for testing a policy before you enforce it, because it works even while enforcement is off.

• Add ?ipDetection=true to POST /v1/generate/custom/{modelId} to run a check on that request and receive the findings alongside the response.

• Combine it with &dryRun=true for an advisory check that returns findings without generating anything.

• A request blocked by enforcement fails with HTTP 422 and a message naming intellectual-property risk as the reason.

Each finding tells you which filter spoke, what it recognized, and how sure it was:

"ipDetection": {
  "action": "blocked",
  "flagged": true,
  "evaluatedFilters": 3,
  "creativeUnitsCharged": 2,
  "findings": [
    {
      "filterName": "Fictional characters",
      "category": "fictional-character",
      "flagged": true,
      "entities": ["Batman"],
      "reason": "The prompt names a specific copyrighted character.",
      "confidence": 0.95
    }
  ]
}

9. Good to Know

• Prompts and input images are screened. Video, audio and 3D inputs are not analyzed, and are not charged.

• An input image that cannot be read counts as a check that did not complete. With Block unchecked jobs on, that job is blocked.

• Up to ten filters run on a single job: the four built-in categories plus your six custom ones.

• IP Detection is a risk signal to help you enforce your own policy. It is not legal advice, and it does not replace the model providers' own content restrictions.