Mimic Party Scoring: How the AI Grades Your Imitation
What the AI actually measures
FoliesApp describes the scoring in the same terms on its Steam page and on its own Mimic Party page. Put together:
| Measured | What it means for your take | Source |
|---|---|---|
| Melody | The shape of the pitch curve, compared in relative pitch — your register never counts against you. | foliesapp.com |
| Rhythm | Timing and length across the whole take. | Steam page, foliesapp.com |
| Attacks | The number of distinct onsets — “whether you barked once or three times”. One bark does not pass for three. | Steam page, foliesapp.com |
| Not measured: timbre | How your voice sounds. “A deep voice and a high voice doing the same curve score the same.” | Steam page |
| Not measured: your microphone | ”Not the microphone you bought”, and not the size of the room you record in. | foliesapp.com |
The analysis runs on your own machine, not in the cloud, while the round keeps going. The score appears right after your take plays, out of 100, in front of everybody.
Scores can go over 100
Wheel bonuses are added on top of the imitation score. One official screenshot shows a take landing at 101 / 100, and players in launch-week videos read out bonuses like “10 points added to your next imitation”. A multiplier (the wheel’s x1.5 segment) scales the next score instead. Details on the wheel page.
Why a “perfect” imitation can score 0
This is the most common complaint in the Steam discussions and in gameplay videos. One player writes that a friend did the sound better but “gain[ed] 0 and me 52 points”, and suggests letting players vote instead. In launch-week sessions, streamers regularly get 0 for takes the room thought were the best of the round.
Given how FoliesApp describes the scorer, the usual reasons are predictable:
- Wrong number of attacks. The scorer counts separate sounds. Three short barks for a one-bark reference (or a smooth “meow” for a two-part one) loses points no matter how convincing it sounds.
- Timing off. Starting late or running long changes the rhythm comparison.
- Doing the voice instead of the shape. A great character impression with a flat pitch curve reads as the wrong melody. The scorer ignores the timbre that makes impressions funny.
- Noise in the take. Laughing, talking or speaker audio in the recording adds onsets the scorer tries to match. Headphones help.
- A sabotage. If someone aimed a sabotage at you, your take can come out echoing, chopped, cut off or replaced entirely — and it costs points on top.
How to score higher
These follow directly from what FoliesApp says is measured; they are not official tips.
- Count the sounds first. On the announce, count how many distinct hits the reference has. Match that number exactly.
- Trace the pitch curve. Rising, falling, wobbling — copy the direction and size of the movement in your own comfortable register.
- Match the length. Start on the countdown and stop when the reference would stop.
- Keep the take clean. Mute your laugh until the playback, use headphones, and don’t talk over your own recording.
- Save the performance for the room. The funniest takes and the highest-scoring takes are often different takes. Pick which game you are playing.
Is the scoring broken?
Players disagree about it, and the Steam review summary sits at Mixed (143 positive, 181 negative out of 324 on September 10, 2026). FoliesApp has not published any change to the scorer since launch. If your group prefers laughs over numbers, one host in a launch-week video simply switches the wheel off and treats the scores as a running joke.
Thanks for the feedback!