Native multimodal output
A single prompt produces matching text, keyframes and video, with consistent characters, style and lighting carrying across formats.
Leaking · Google I/O 2026 Gemini Omni
Gemini Omni
One model for text, image, video and audio
Surfacing in early May 2026 across multiple leaks, Gemini Omni is Google's upcoming unified multimodal model: native generation of text, image, video and synced audio inside a single Gemini-trained system.
Unified model Synced audio In-chat editing
Omni
Text
Image
Video
Audio
Quick stats
5–10s Clip length
1080p Max output
16:9 · 9:16 · 1:1 Aspect ratios
I/O 2026 Expected reveal
Capabilities
The whole pipeline collapses into one model
Unlike specialised video models such as Veo, Sora 2, Seedance 2.0 or Kling, Gemini Omni keeps language reasoning, image generation, video generation and audio synthesis under one architecture.
One unified Gemini stack
No more chaining of specialised models. Text, image, video and audio share the same weights and the same long context.
Synced native audio
Ambient sound, score and dialogue are aligned with the picture in the same forward pass — footsteps land on the beat, lips match speech on first export.
Direct in-chat editing
Swap an object, change the lighting, adjust a camera move in natural language — no full regeneration, echoing the Nano Banana editing playbook.
Remix and steer
Upload an existing clip and redirect it with prompts. Reference images, videos and audio can be combined in a single instruction.
Templates & styles
Built-in templates for product ads, Reels, music videos and cinematic shorts lower the floor for first-time users while keeping camera language consistent.
Specs
What can be pieced together before the keynote
Numbers below are aggregated from Reddit/X leaks and reporting by TestingCatalog, Programming Insider and OfficeChai.
| Dimension | Known signal |
|---|---|
| Model family | Google Gemini — successor branding for the Veo line |
| Model ID | bard_eac_video_generation_omni / v3smm-lora-prod |
| Clip length | 5 / 8 / 10 seconds per generation, chainable in-app |
| Resolution | 480p / 720p / 1080p |
| Aspect ratios | 16:9, 9:16, 1:1 |
| Audio | Natively synthesized, synced in a single pass |
| Inputs | Text / image / video / audio references |
| Access | Staging inside Gemini app, API expected post I/O |
| Quota signal | Reports say two Omni generations burn ~86% of an AI Pro daily quota |
Architecture
Three product lines collapse into one Omni
Google's generative stack used to be split across Veo for video, Nano Banana / Imagen for image and Gemini for text. Omni rolls those into a single architecture.
Before
Veo 3.1
Video + native audio
Nano Banana / Imagen
Image generation & editing
Gemini 2.5 / 3.x
Reasoning · long context
Now · Omni
Gemini Omni
Text · image · video · audio, one model, one prompt
Text Image Video Audio
Use cases
From a single brief to publishable content
A unified model with long context and synced audio means teams can write one coherent brief and walk away with a finished cut.
01
Product ads
Hero shots, packaging reveals and lifestyle cuts shipped with ambient audio already locked.
02
Reels & Shorts
Vertical 9:16 clips with on-mic dialogue and beat-synced motion, built for scroll-stopping social.
03
Music videos
Reference a track and Omni cuts visuals to the beat, keeping a consistent character across shots.
04
Cinematic shorts
Chain multiple 10-second omni-clips into multi-shot sequences with continuous lighting and audio bed.
05
Landing-page hero loops
Loopable 16:9 atmospheric clips for SaaS, fashion and DTC sites — branded and silent-friendly.
06
Explainers & tutorials
Turn a script into a narrated sequence with lip-synced dialogue and matching ambient sound.
Compare
Where Omni sits in the 2026 video stack
Aggregated from Artificial Analysis, Looksy AI, Oimi AI and the official keynotes — for orientation, not benchmark scores.
| Model | Maker | Architecture | Native audio | Clip length |
|---|---|---|---|---|
| Gemini Omni
Omni
| Unified omni (video + image + audio) | Synced in one pass | 5 / 8 / 10s | |
| Veo 3.1 | Specialised video model | Yes | ~8s | |
| Seedance 2.0 | ByteDance | Specialised multi-modal video | Yes | up to 15s / shot |
| Sora 2 | OpenAI | Specialised video model | Yes | ~20s |
| Kling V3.0 | Kuaishou | Specialised video model | Limited | ~10s |
Timeline
From the first leak to the I/O 2026 stage
Ordered by public report date, still evolving.
- 2026 · 05 · 02
First "Powered by Omni" string
X user @Thomas16937378 spotted "Start with an idea or try a template. Powered by Omni." inside the Gemini video tab.
- 2026 · 05 · 11
Full preview card inside Gemini mobile
TestingCatalog and Chetaslua surfaced the "Meet our new video model" card, the full model ID and the 10-second clip cap.
- 2026 · 05 · 12 – 18
Demos circulate in the wild
A "professor solving trig on a chalkboard" clip showcased text coherence and physical fidelity, sparking heavy comparison with Veo 3.1.
- 2026 · 05 · 19 – 20
Expected unveil at Google I/O 2026
Main-stage time is widely expected for Omni, possibly alongside Flash / Pro tiering, an API, and reshuffled subscription tiers.
FAQ
The questions people ask most about Gemini Omni
What exactly is Gemini Omni?
It's Google's upcoming unified multimodal model that natively generates text, image, video and synced audio inside one architecture — effectively merging Veo, Imagen and Gemini.
When will it ship?
As of mid-May 2026 Omni is still in the leak phase. The widely expected reveal is the Google I/O 2026 main stage on May 19–20.
How does it relate to Veo 3.1?
Metadata suggests Omni inherits engineering from the Veo stack, but it drops the Veo brand and folds video into Gemini's text and image layers.
Does it really generate sound?
Yes. Ambient sound, score and dialogue are produced in the same pass as the video — that's the whole reason for the 'omni' name.
What is the current clip-length limit?
The leaked model ID points to 5, 8 or 10 seconds per generation, with multi-clip chaining at the client layer.
How will pricing work?
Unconfirmed. A Reddit screenshot shows two Omni generations burning ~86% of the AI Pro daily quota, so a higher 'Ultra / Pro Plus' tier is plausible.
Sources
Primary reports and public links
Everything on this page is aggregated from the public sources below. Cross-reading is recommended.
programminginsider.com Read source
TestingCatalog · Programming Insider report
Leak details, UI strings and early demo analysis.
oimi.ai Read source
Oimi AI · Gemini Omni Leaked roundup
Speculation on architecture and side-by-side with Seedance / Veo.
officechai.com Read source
OfficeChai · Gemini Omni Spotted
Full model ID, in-app prompts and community reactions.
looksyai.in Read source
Looksy AI · Gemini Omni product page
Tidy summary of specs, use cases and comparisons.
storage.googleapis.com Read source
Gemini 2.5 technical report
Family-level multimodality, long context and the agentic direction.