Alani · Emotion
Alani
Emotion
Upload or record a short video of a person and see how it reads emotionally, with the evidence behind the estimate.
Opens https://emotion.alatomy.com. You need the access password from Alatomy.
What it is
Reading emotion from video
Alani Emotion watches and listens to a short clip and places the person's emotional state on a simple map with two axes: valence (unpleasant to pleasant) and arousal (calm to energised). The answer is one of four quadrants.
It is built on research into emotion recognition in video calls. It is a research tool and a demonstration, not a medical or diagnostic product.
Positive · Active
Pleasant and energised.
Negative · Active
Unpleasant and energised.
Positive · Deactivated
Pleasant and calm.
Negative · Deactivated
Unpleasant and subdued.
How to use it
Six steps
Open the board and log in
Press the button above and enter the access password you were given.
Warm up (optional)
Press “Warm up models” if the badge says they are not loaded. The first analysis after a quiet period is slower, and this gets the models ready.
Add a video
Drop a video file in the box, or press “Record with camera” to record straight from your device. Short clips of 30 to 60 seconds are plenty.
Choose what to use
Pick which signals the analysis may use (see below), or use a preset. Leave “Retry automatically” ticked so a busy moment on the model side does not stop your run.
Watch it work
A progress bar and a plain-language description show each step as it happens, with how long it took.
Read the results
See the predicted quadrant, the annotated frame, the evidence, and the exact questions the model was asked. Try other signals and compare, or send the result to Discord.
What it looks at
Five signals
AU · Facial action units
Which facial muscle movements are present, such as a raised cheek or a lowered brow.
T · Transcript
What the person says, turned into text.
VO · Voice
How the voice sounds: pitch, loudness and speaking rate.
ET · Eye gaze
Where the person is looking, and how steadily.
HT · Head pose
How the head is turned, tilted and moving.
The video itself
Sampled frames from the clip are always shown to the model alongside the chosen signals.
The results page
What you get
- The quadrant, with a colour and a shape, plotted on the valence and arousal map. The estimate is a quadrant, not a score.
- The annotated frame: the face with landmarks, action-unit bars and head pose drawn on it, plus a filmstrip of the frames used.
- The evidence: the strongest action units, head and gaze dials, the transcript and the voice summary.
- What the model saw: both questions it was asked and its one-word answers.
- Timing: a colour-coded bar of how long each step took, and which steps were reused.
- Try other signals: run the same clip again with a different set and compare the runs side by side.
Good to know
Limits and privacy
First two minutes
Only the first 2:00 of a clip is analysed. The page tells you when a clip was cut.
Not a diagnosis
An estimate from one short clip. Treat it as a research demonstration.
Your video is deleted
Uploads are deleted automatically after 24 hours unless you pin the run, and you can delete a run at any time.
It can be busy
Analyses run one at a time, so you may wait in line. The page shows your place.