Compare real render outputs
Choose one of your six existing renders, or select a video from your device. These are completed videos, not live editable avatars. Uploaded selections clear on reload.
B is muted to prevent overlapping audio. Side-by-side playback is approximate, not frame-locked.
Automatic file readings
Available from completed videos. No new GPU render needed. File size and resolution do not prove facial detail or lip-sync quality.
Select videos to see readings.
Facial diagnostics · local analysis
Measure sampled face visibility, sharpness proxy, brightness, face-center movement and eye/mouth openness variation. Up to 40 frames per clip, maximum 60 seconds. MediaPipe runs on this device; video frames are not uploaded.
Not analyzed yet.
Sharpness is measured on a resized 128×128 face crop. Compare similar identity, framing and lighting. Higher sharpness may reflect noise or oversharpening. Sparse samples can miss blinks and brief defects. Lip-sync accuracy and identity drift require separate validated models and are not scored here.Which result should we build on?
Watch both with the same spoken audio and comparable framing. Select a preference for each criterion; these are your observations, not AI-generated scores.
No quality decision yet. Readings alone cannot choose a winner.
The summary counts preferences equally; any unacceptable criterion blocks a ready-to-use recommendation. Save test record below to keep readings and observations.Save an evaluation
Scores are human-entered. Render cost covers entered render seconds only, not startup, idle time or storage.