/Tech34d ago

AI experimenter fofr finds Gemini Omni struggles with two-handed finger-counting in video evaluation

The model failed specifically on numbers seven and eight

35258175443.3K

#904

Original post

fofr@fofrAI#1853inTech

One year later with Omni and this test can pass. I saw it getting pretty close, so I tweaked the prompt:

> A video of a man counting to 10 on his fingers, show the number in the corner. A new number every 1s, no dialogue other than the numbers he says. He uses two hands for numbers bigger than 5.

- the model does 1 to 5 consistently well - struggles more when two hands are used, usually on 7 and 8 - if you ask it to count faster, errors increase - it keeps a good cadence

fofr@fofrAI

A new prompt to add to the fofr-benchmark:

> a man counts out loud from 1 to 10, using his fingers and holding them up as he goes

> a man counts out loud from 1 to 10, "1, 2, 3, 4, 6, 7, 8, 9, 10", he counts using his fingers and holds them up as he goes

4:10 AM · May 26, 2026 · 32K Views

Sentiment

Positive users praise Omni AI video model's finger-counting benchmark gains as impressive progress in generation quality, while negative users criticize remaining flaws like hand switching and question readiness for serious work.

Pos

77.8%

Neg

22.2%

12 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS4.1KBOOKMARKS7LIKES36RETWEETS3REPLIES5

fofr@fofrAI

1 to 20 also works pretty well, if we're just counting

fofr@fofrAI

One year later with Omni and this test can pass. I saw it getting pretty close, so I tweaked the prompt:

> A video of a man counting to 10 on his fingers, show the number in the corner. A new number every 1s, no dialogue other than the numbers he says. He uses two hands for numbers bigger than 5.

- the model does 1 to 5 consistently well - struggles more when two hands are used, usually on 7 and 8 - if you ask it to count faster, errors increase - it keeps a good cadence

34d4.1K367

fofr@fofrAI

Time to make the test harder

> A video of a man counting as fast as he can, show the number in the corner. No dialogue other than the numbers he says.

fofr@fofrAI

One year later with Omni and this test can pass. I saw it getting pretty close, so I tweaked the prompt:

> A video of a man counting to 10 on his fingers, show the number in the corner. A new number every 1s, no dialogue other than the numbers he says. He uses two hands for numbers bigger than 5.

- the model does 1 to 5 consistently well - struggles more when two hands are used, usually on 7 and 8 - if you ask it to count faster, errors increase - it keeps a good cadence

34d3K356

Emily@IamEmily2050

Omni Flash is so much better than Veo3 and smarter than Seedance V2, but there are two problems at the moment. First, the censorship when it comes to anything cinematic, dramatic, or anime. Second, it's not native 1080 and just ten seconds. You can see from the video, and I am sure the team is more than aware of these things.

34d423123

Yash Bhalgat@ysbhalgat

hahah this is one of the toughest tests for video-gen models.

historically, video generation models have struggled with counting and in general with "numbers". really glad to see Gemini Omni doing so well on this task!

fofr@fofrAI

One year later with Omni and this test can pass. I saw it getting pretty close, so I tweaked the prompt:

> A video of a man counting to 10 on his fingers, show the number in the corner. A new number every 1s, no dialogue other than the numbers he says. He uses two hands for numbers bigger than 5.

- the model does 1 to 5 consistently well - struggles more when two hands are used, usually on 7 and 8 - if you ask it to count faster, errors increase - it keeps a good cadence

34d2K103

fofr@fofrAI

With the exact same prompt from last year

fofr@fofrAI

One year later with Omni and this test can pass. I saw it getting pretty close, so I tweaked the prompt:

> A video of a man counting to 10 on his fingers, show the number in the corner. A new number every 1s, no dialogue other than the numbers he says. He uses two hands for numbers bigger than 5.

- the model does 1 to 5 consistently well - struggles more when two hands are used, usually on 7 and 8 - if you ask it to count faster, errors increase - it keeps a good cadence

34d1.6K123

fofr@fofrAI

Other models for reference, Seedance 2:

34d1.4K71

Rambam the grey@meowbooksj

@fofrAI mfw 7

34d1275

fofr@fofrAI

Sora 2:

34d7583

fofr@fofrAI

@madpencil_ Nice idea. It can do a countdown from 10 consistently correctly. But it fails with a countdown from 20.

34d572

fofr@fofrAI

@giffmana 😅

Lucas Beyer (bl16)@giffmana

@fofrAI Of course you did it already, hours before my question!

34d60620

fofr@fofrAI

@meowbooksj Yeah, switching hands like that threw me.

34d863

gokaygokay@gokayfem

@fofrAI finally 🔥

34d1162

Mark Sabalauskas@MarkSab

@fofrAI Neat. I asked it to build a memory palace counting up from 1-5 in Mandarin, showing me the Traditional Chinese character, speaking the word in Mandarin, showing me how to pronounce it, with a visual cue to remember the pronunciation.

34d1041

Philipp@jphilipp

@fofrAI Checking if it can count in multiple languages in one video.

34d152

Rishu Kumar@rishdotuk

@fofrAI Switching hands like that should be declared a crime against humanity in the Geneva Convention.

34d451

madpencil_@madpencil_

@fofrAI A video of an young adult rapping counting as fast as he can, his attire flicker changes each number he counts, high speed rap, show the number in the corner. No words other than the numbers he says.

Cool 😄

34d90