Reality Defender's video detection analyzes content using multiple specialized models to identify signs of manipulation. Each video receives a single overall judgement and score, produced by combining signals from all models that ran on that file.
What's included in video analysis
Video analysis is separated into two model groups: full-frame video models and face-region video models. Applicable signals from these models are combined by rd-vid-ensemble into one overall video result.
Full-frame video models
Full-frame models analyze the complete video frame, rather than only detected faces. These models expand coverage for videos where manipulation signals may appear outside a face, or where faces are absent, small, blurry, heavily compressed, partially occluded, or not consistently detectable.
rd-full-dynamics-vid (Full-frame Dynamics) — analyzes motion and temporal patterns across the complete video frame.
rd-full-erie-vid (Full-frame Universal) — analyzes the complete video frame for manipulation signals that may appear in the body, background, scene, or surrounding visual context.
Face-region video models
Face-region models analyze detected faces when usable face regions are available.
rd-dynamics-vid (Dynamics) — analyzes temporal inconsistencies and motion dynamics across detected face regions.
rd-erie-vid (Universal) — analyzes facial and motion-based manipulation signals across frames.
rd-tahoe-vid (Guided) — retained in the API response for backwards compatibility, but this model has been deprecated.
Ensemble model
rd-vid-ensemble — combines applicable full-frame and face-region model signals into a single calibrated assessment of manipulation likelihood for the file
What you'll see in the API response
The API response includes an overall result object with the file-level status and score. This is the final conclusion on the video file.
Each model returns its own result in the models[] array, including the model name, status, and score.
Models that are not applicable to a given media type (for example, image or audio models on a video file) will return "status": "NOT_APPLICABLE" with null score fields. This is expected behavior.
How to interpret video model results
Individual model scores reflect each model's independent assessment. Because each model is trained to detect different features of manipulation, it is normal for them to diverge. The ensemble accounts for this, weighting each model's output to produce a single calibrated score.
You may see cases where:
Full-frame models return high scores — manipulation signals may be present in the broader frame, motion, scene, body, background, or surrounding context
Face-region models return high scores — manipulation evidence may be concentrated in detected facial features
Full-frame models trigger while face-region models do not — the video may not contain a usable face, or relevant signals may appear outside detected faces
Face-region models trigger while full-frame models do not — the manipulation signal may be localized to the face region
Some models return
"NOT_EVALUATED"— the model did not produce a scoring result for that file or detection pathIndividual model scores are mixed, but
rd-vid-ensembleis elevated — the combined weight of evidence still indicates a meaningful signal
For most use cases, the overall video conclusion and score should be used as the primary decision signal. Individual model results are available for debugging, auditing, or deeper investigation.
Context-aware analysis
Video results may also include a signal from rd-context-vid, which applies contextual analysis to the video beyond frame-level detection. When present, this signal is factored into the overall result. When not applicable, it will appear with "status": "NOT_APPLICABLE".