Video Deepfake Detection
Response schema, prediction fields, and example outputs for AI-ForensiX video deepfake detection model.
AI-ForensiX Video Deepfake Detection model analyzes multiple frames over time to determine whether a video is manipulated, reenacted, AI-generated, or authentic.
It provides classification labels, confidence scores, XAI heatmaps, and manipulation source identification.
Response Schema
VideoDeepfakeDetectionResult
| Field | Type | Description |
|---|---|---|
label | string ("real" | "fake") | Predicted classification indicating whether the video is authentic or manipulated. |
score | number (0.0 – 1.0) | Confidence score aggregated across the entire video using temporal frame analysis. |
heatmap_url | string (URL) - optional | URL to heatmap visualization showing temporal + spatial evidence of manipulation. |
source | string ("real" | "face_reenactment" | "face_edit" | "ai_generated") | Indicates the manipulation type or authenticity. |
Source Classification Explanation
| Value | Meaning |
|---|---|
| real | Video is authentic and unaltered. |
| face_reenactment | Facial motion-transfer detected (expression puppeteering). |
| face_edit | Facial attribute editing/enhancement detected in video frames. |
| ai_generated | Entire video or major segments generated by AI. |
Example Responses
Listing: Fake Video Detection Example
{
"label": "fake",
"score": 0.893,
"heatmap_url": "https:https://forensiX.com/.mp4",
"source": "face_reenactment"
}Listing: Real Video Detection Example
{
"label": "real",
"score": 0.975,
"source": "real"
}