EmoteBench · Keyframe Research
A new standard for expressive avatars
Avatar benchmarks measure lip sync, latency, and preference. EmoteBench measures emotional expression. When the sentiment is grave we expect to see concern; when it brightens, we expect the face to brighten with it.
EmoteBench is a blinded human study of whether real-time avatars do that. 250 raters made 19,500 judgments across five avatar APIs, rating whether the face expressed the right emotion, looked natural, and changed when prompted. Keyframe scored highest on every measure, against every competitor.
Expression
Named the emotion on Keyframe's face
95% CI 76–85%
Naturalness
Called Keyframe's face very or completely natural
95% CI 54–67%
Control
Saw Keyframe's face switch emotion when prompted and named both emotions
95% CI 41–64%
Expression
Raters could name the emotion on Keyframe's faces more than twice as often as on any other.
Shown a face , raters picked which of seven emotions it showed. Keyframe was read correctly 81% of the time. Anam, Tavus, and LemonSlice landed between 32% and 37%, where guessing scores 14%.
Share of silent clips where raters chose the intended emotion. About 2,250 ratings per system; the dashed line is chance.
| Emotion | Keyframe | Anam | Tavus | LemonSlice |
|---|---|---|---|---|
| Anger | 97% | 22% | 41% | 67% |
| Happiness | 95% | 49% | 94% | 41% |
| Neutral | 87% | 77% | 30% | 76% |
| Sadness | 84% | 16% | 45% | 8% |
| Fear | 79% | 2% | 12% | 4% |
| Surprise | 69% | 25% | 30% | 15% |
| Disgust | 58% | 49% | 13% | 10% |
Keyframe has the highest share on all seven emotions, though Tavus is within a point on happiness. On fear, the three other systems score between 2% and 12%.
Naturalness
More expressive, and more natural.
With the sound on, raters also judged Keyframe both more natural and clearer in expressing the emotion than every other system, and every one of Keyframe's faces outranked every competitor's.
Looked natural
Showed the emotion clearly
Share of ratings at the top, middle, and bottom of a five-point scale. About 1,300 ratings per system; HeyGen, which has no emotion control, was rated on neutral and happy lines only.
Mean naturalness, 1 to 5, for all fifteen faces. Keyframe's lowest, Caspian at 3.42, sits above the best from any other provider, Anam's Mia at 3.03.
Control
The most steerable face of any model tested.
Raters watched silent clips prompted to switch emotion mid-sentence and had to name the emotion before and after. On Keyframe they named both correctly 53% of the time; on Tavus 13%, LemonSlice 5%, Anam 3%. Prompted to hold instead, Keyframe's face kept its expression 86% of the time, while Tavus, LemonSlice, and Anam drifted off theirs on 28% to 41% of clips. Raters could barely tell whether those faces had been told to hold or to switch.
Solid: share of prompted switches where raters saw the change and named both emotions; chance is 2%. Hatched: share of control clips, prompted to hold, where raters reported a switch anyway.
Head to head
A large majority of raters put Keyframe ahead of every competitor.
Nearly every rater saw every system. For each of the 250 raters we compared their average score for Keyframe against their average score for each competitor. On recognizing the emotion, 98% of raters put Keyframe ahead; on showing it clearly, 93% to 97%; on looking natural, 77% to 87%. Even the closest comparison, naturalness against Anam, went Keyframe's way for more than three raters in four. Rating against rating on the same line, Keyframe still wins 69% to 87% of the time.
Emotion recognized
Emotion shown clearly
Looked natural
One square per rater who saw both systems, 246 to 250 per comparison. Line by line, Keyframe also won 88% to 100% of the lines both systems performed. Every lead's 95% interval excludes zero, both in all sessions and in the half that passed attention checks. HeyGen has no emotion control and was not tested on recognition.
Try it
Judge the faces yourself.
Real screens from the study. Pick an answer, then see what the raters chose. Clips play as raters saw them, one system behind each letter. The full report has every breakdown, interval, and example line.
Behind the Expression result above · No sound. Name the emotion the face is showing.
Which emotion does the face show?
No sound on this one. Watch the whole clip, then answer.
The face is intended to show surprise
Sound on. Rate the face only, not the voice.
Emotion switching
No sound on this one. Some faces switch emotion part-way through and some don't.
The study
How it was measured
Every system received identical audio for every line, so the only thing that differed between two clips was the face. 182 scripted lines across seven emotions, three stock identities per provider, clips captured live from each API and rated blind.
Late October 2026
Clips captured live from each provider's streaming session, on the versions available at the time.
250 raters · 19,500 rated clips
An online research panel. One clip per screen, the system shown as a letter, no line seen twice.
Five real-time avatar APIs
Keyframe persona-2.0-live, Anam cara-4.1-express, Tavus Phoenix-4.5, LemonSlice 2.1, HeyGen LiveAvatar.
Keyframe designed and ran the study
Raters never saw provider names. Every system is reported on every measure it could take.
Code and data: coming soon