Google Gemini Omni
Google Gemini Omni is a multimodal model released in May 2026 that includes video generation, and is reported to exceed Sora 2 on the capabilities that mattered most to its users.
- Best category
- Lip-sync 4.99
- Weakest category
- Hands 4.01
- Clips scored
- 40
What it is actually for
Gemini Omni landed on 19 May 2026 as a multimodal model rather than a video model, with video generation as one capability among several. Reporting at the time put it ahead of Sora 2 on the capabilities Sora users cared about most — which, given Sora was already on its way out, made it a natural landing spot.
Why it is listed separately from Veo
Because they are different products with different shapes. Veo 3.1 is a dedicated video model you call to make a clip. Omni is a general multimodal system where video is one output among several, which matters if your pipeline is doing more than one kind of generation and you would rather not stitch together three vendors.
What we do not know yet
Almost everything, honestly. It is the newest entry we track and we have not put it through the battery. We have listed it because leaving a relevant model off a comparison is worse than listing it with an explicit gap where the numbers should be.
Capabilities
- Native audio
- Yes
- Multi-shot sequences
- No
- Image to video
- Yes
- Public API
- Yes
Benchmark scores
How we test →| Dimension | Score | |
|---|---|---|
| Overall | 4.82 / 5 | |
| Prompt fidelity | 4.81 | |
| Visual quality | 4.87 | |
| Physics & motion | 4.80 | |
| Subject consistency | 4.81 | |
| Audio & sync | 4.82 |
- Scored runs
- 40
| Prompt category | Score | Runs | |
|---|---|---|---|
| Lip-sync | 4.99 | 4 | |
| Faces | 4.96 | 4 | |
| Multi-shot | 4.96 | 4 | |
| Camera | 4.96 | 4 | |
| Style | 4.96 | 4 | |
| Image to video | 4.96 | 4 | |
| On-screen text | 4.91 | 4 | |
| Animation | 4.85 | 4 | |
| Liquids | 4.79 | 4 | |
| Animals | 4.73 | 4 | |
| Hands | 4.01 | 4 |
Scores out of 5, judged blind. Bars start at 3.0 so the gaps are visible.
Cite this
Free to cite and reuse with a link back. Data is updated as new runs and prices come in, so include the date.
SlopTV. (2026). Google Gemini Omni benchmark results. Retrieved October 23, 2026, from https://sloptv.co/model/gemini-omni<a href="https://sloptv.co/model/gemini-omni">Google Gemini Omni benchmark results</a> (SlopTV)The clips
Every score above links to the actual generation it came from. Nothing here is a mockup.
Hands avg 4.01/5
Chopping a pepper
Clean grip and contact, but half of the pepper vanishes after the first cut.
Counting fingers
Five digits and normal anatomy, but the hand flips over midway through the clip.
Fretting a guitar chord
Fingertips press the strings cleanly, but the shot cuts twice to a new framing during what should be one close-up.
Tying shoelaces
Knot shape morphs slightly as it tightens, otherwise clean.
Faces avg 4.96/5
Liquids avg 4.79/5
Honey drizzle
Dipper visibly starts and stops the pour -- no endless/sourceless stream, unlike the same prompt on Runway Gen-4.5 and PixVerse 5.5.
Ice melting in whisky
Exterior condensation droplets appear to materialize unnaturally fast on the glass.
Wave on rocks
Abrupt shot change around 00:04.5 not requested by the prompt; fine spray shows slight generative softness.
On-screen text avg 4.91/5
Book chapter heading
Heading 'CHAPTER ONE' correct; surrounding body text is expected filler gibberish.
Multi-shot avg 4.96/5
Cafe, three shots
Real cuts (wide -> medium -> close-up) with fully consistent identity, coat, mug and cafe across all three -- the intended multi-shot execution.
Same bench, day and dusk
Real cut with an actual midday-to-dusk lighting change (lamps lit, atmosphere shifted) -- the only one of the models benchmarked so far to render this prompt as intended rather than a single static/continuous shot.
Watch, wide then macro
Real cuts from full watch to macro dial to internal movement, fully consistent.
Alley, two angles
Real cut from the rear tracking shot to the frontal shot as he turns the corner, fully consistent.
Lip-sync avg 4.99/5
Bearded man, long line
Speaks the requested line accurately with natural, precisely synced articulation.
Direct address to camera
Speaks the requested line accurately with tightly synced mouth movement, then repeats a fragment of it ('you do not have to') a second time -- minor script deviation, everything else clean. Real, accurate speech audio confirmed.
Sung line
Sings the requested lyric with mouth shapes convincingly synced to the sustained vocal notes.
Two-person exchange
Both speakers deliver their exact lines with tight, natural lip-sync.
Camera avg 4.96/5
360 orbit
Completes a full 360-degree orbit (passes behind the subject and returns to front) -- the only model so far to fully execute this prompt rather than a partial arc.
Animals avg 4.73/5
Pigeon takeoff
Bird actually leaves the ledge and completes a convincing flight with correct wing/tail mechanics -- unlike Runway Gen-4.5 and PixVerse 5.5, where it never took off.
Cat jumping to a counter
Mildly floaty apex with slight torso elongation mid-air; landing and subsequent behavior are convincing.
Galloping horse
Gait looks somewhat rigid with foot-sliding and lacks natural vertical heave/ground-impact compression.
Style avg 4.96/5
Rotating mug
Completes a full 360-degree rotation with the handle disappearing and reappearing correctly -- unlike PixVerse 5.5, where the mug barely rotated.
Image to video avg 4.96/5
Animation avg 4.85/5
3D feature film look
Edited as three shots: a wide of the robot sitting in a forest, a close-up of its eyes widening and an insert of the butterfly landing on its fingertip with a small glow. The character design matches across the cuts.
Anime sword draw
A slow, tense draw for three seconds, a slash with impact lines, then cuts to a side-view lunge and a back view with hair trailing. Strong follow-through, but not a static camera.
Squash and stretch
Textbook cartoon timing: anticipation before the hop, a flat squash on landing, a stretched rebound with speed lines and a second squash before the surprised face. The camera pans down to follow the fall.
2D walk cycle
A clean four-legged walk in side view with an even rhythm, the head bobbing with each step and the scarf trailing. Design unchanged from start to end.
Where to use it
The same model costs different amounts depending on the platform you reach it through.
| Platform | 10s costs | Max res | Watermark | Commercial use | |
|---|---|---|---|---|---|
 Magnific |
— | 4K | no | yes | Visit |
 Leonardo.Ai |
— | — | — | — | Visit |
 Krea |
— | — | — | yes | Visit |
 OpenArt |
— | 4K | no | yes | Visit |
 Lumeta |
— | — | no | yes | Visit |