> For the complete documentation index, see [llms.txt](https://docs.lassomoderation.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.lassomoderation.com/content-analysis/ai-image-detection.md).

# AI Image Detection

Detect images and videos that declare themselves as AI generated in their metadata, and use that in your moderation rules.

AI image detection reads what an image or video file says about its own origin. Many AI tools record in the file that it was generated, either as signed Content Credentials or as ordinary metadata. Lasso reads that information when it processes an image or video, shows it to your moderators and lets you use it in rules. For example, you can flag AI generated profile photos for review, or tag every image that declares it was made with AI.

{% hint style="info" %}
This feature finds images and videos that **declare** themselves as AI generated. It does not analyze the pixels. Metadata is often lost when a file is shared, screenshotted or re-encoded, so no match does **not** mean an image or video is real. Images and videos are two separate features: **AI image detection** and **AI video detection**. Contact Lasso to enable either for your workspace.
{% endhint %}

### What you can use it for

* **Label AI media**: tag images and videos that declare AI, so moderators can see them and you can measure how common they are.
* **Enforce an AI image policy**: flag AI images in places where your policy does not allow them, such as marketplace listings.
* **Prioritize risky combinations**: combine it with other conditions to review AI images that also contain a recognized public figure, or AI generated profile photos.

### What Lasso checks

Lasso looks for the following, from strongest to weakest evidence:

* **Content Credentials (C2PA)**: a signed record of how the image was made. Lasso checks the signature and whether the image was changed after signing.
* **IPTC digital source type**: a standard metadata field that marks an image as generated by a trained AI model.
* **Generator names**: the name of a known AI tool in the file's XMP or EXIF software fields.
* **PNG generation data**: the settings that tools such as Stable Diffusion interfaces and ComfyUI store in PNG files.

Only Content Credentials can be verified. The other sources are plain text that anyone can edit, so treat them as a strong hint rather than proof.

An image counts as **fully AI generated** when the image was created by an AI model. It counts as **partly AI generated** when AI was used on part of it, for example a real photo with one AI edit.

### Videos

AI video detection reads the original video file before Lasso takes screenshots or extracts audio. It checks:

* **Content Credentials (C2PA)** in MP4, MOV and AVI files, with the same signature check as for images.
* **IPTC digital source type** and **generator names** in the video's XMP metadata.
* **Generator names** in the video's own metadata tags, such as its comment or encoder.

It works the same way as for images, with its own condition: **AI generated video** (see below). Video screenshots are not checked separately, because a screenshot never carries the original file's metadata. Videos lose their metadata even more often than images: most apps and social platforms re-encode uploaded videos, which removes it.

### Using it in rules

Add the **AI generated image** condition to a content rule or a queue rule, or **AI generated video** for videos. See [Custom Moderation Rules](https://docs.lassomoderation.com/automation-and-review/custom-moderation-rules.md) for how rules work. Both conditions have the same two options:

| Option           | Choices                                   | Meaning                                                                                                                      |
| ---------------- | ----------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------- |
| What it declares | **declares fully AI generated**           | Only images and videos created by an AI model                                                                                |
|                  | **declares fully or partly AI generated** | Also includes media where AI was used on part of it                                                                          |
| Evidence         | **any metadata**                          | Any declaration, signed or not. Matches the most media                                                                       |
|                  | **a verified signature**                  | Only Content Credentials with a valid signature from a certificate on a trusted list. Harder to fake, but matches less media |

**AI generated image** matches when any image in the content matches, **AI generated video** when any video does.

<figure><img src="https://2736176226-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FFslADZ1R9paAfD3FHscR%2Fuploads%2Fgit-blob-6a204b76beedff256393640069db3e73a89aa313%2Fai-generated-image-condition.png?alt=media" alt="The AI generated image condition in the rule editor, with its two dropdowns and help icon"><figcaption></figcaption></figure>

We recommend these settings:

* Use **any metadata** for tagging and sending content to a review queue.
* Use **a verified signature** with **declares fully AI generated** for automatic actions, and only if your policy does not allow AI media at all.

Do not rely on this condition alone to enforce a ban on AI media, because anyone can remove the metadata.

### Seeing it on content

When an image or video declares AI, the content's details show an **AI detection** row. It lists the declaration, the generator if it is known, and whether the signature was verified, for example:

> Declares fully AI generated · Acme Image Generator · Not verified

Hover over the row to see what each part means. When several images or videos declare AI, the row shows the strongest declaration. Lasso does not show the row for content without a declaration.

<figure><img src="https://2736176226-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FFslADZ1R9paAfD3FHscR%2Fuploads%2Fgit-blob-9cbaaeb0ab51b8b7d1d1f94d014545b8a65ab148%2Fai-detection-content-details.png?alt=media" alt="An AI generated image of a dog with its details below. The AI detection row shows Declares fully AI generated, OpenAI and Verified signature"><figcaption></figcaption></figure>

### What it does not cover

* **Video screenshots**: screenshots taken from a video are not checked. The video file itself is, when AI video detection is enabled.
* **Linked videos**: YouTube and Vimeo links are not processed, so they are not checked.
* **Older content**: only images and videos processed after the feature was enabled have this information. The condition does not match older content.
* **Credentials stored elsewhere**: some files point to Content Credentials hosted on another server. Lasso does not fetch these, so they do not match unless the file also declares AI in its other metadata.
