Gemini Omni: Google's any-to-any model for video — text, image, audio, video input, video output

Gemini Omni: Google's any-to-any model for video — text, image, audio, video input, video output

🔎 Google just changed the rules of video generation

On May 19, 2026, at Google I/O, the company didn't just release a new model. It redefined what "multimodal" means. Gemini Omni arrives as a family of any-to-any models: you give it text, an image, an audio clip, or an existing video, and it returns a video with synchronized audio. A single model to do it all.

This is a paradigm shift. Until now, video generation required separate pipelines: one model for the image, another for motion, a third for sound. Omni merges all of that. And the first public model, Gemini Omni Flash, is already available in the Gemini app, Google Flow, and YouTube Shorts.

Why now? Competitive pressure is at its peak. Seedance 2.0 de Bytedance has been dominating the video generation rankings for weeks. OpenAI's Sora 2 is starting to go mainstream. And Google's own Veo 3.1 models, while highly performant, remain confined to a classic text-to-video approach. Omni is Google's answer to this market fragmentation.

The essentials

• Gemini Omni is a new family of AI models announced on May 19, 2026, at Google I/O, designed as an any-to-any "world model".
• Omni Flash, the first public model, generates 10-second videos from any combination of inputs (text, image, audio, video).
• Conversational editing allows you to modify a video by chatting with the model, without starting from scratch.
• Available immediately in the Gemini app, Google Flow and YouTube Shorts for paying users, with API access planned for later.
• Omni natively integrates synchronized audio generation, which sets it apart from video models that require a separate audio tool.
• Positioned as a direct rival to Sora 2, Veo 3.1 and Seedance 2.0, with the advantage of the Google ecosystem.

Recommended tools

What Gemini Omni really is — and what it is not

Gemini Omni is not just a simple upgrade to Veo. It is a different architecture, designed from the ground up as a "world model" that understands the relationships between all types of media.

According to Google's official announcement, Omni leverages a level of multimodal understanding "grounded in reality." Concretely, the model doesn't just pixelate frames. It simulates physical interactions, light reactions, and coherent sound behaviors.

What fundamentally sets it apart from Veo 3.1 is the nature of the accepted inputs. Veo 3.1, even in its audio variant, primarily works as text-to-video. Omni accepts any combination: a photo + a voice description + a musical excerpt, and it produces a unified video. This is what 9to5Google describes as the ability to "create anything from any input".

Nor is Omni a generalist model like Gemini 3.1 Pro. It is a model specialized in the creation and editing of video media, with an understanding of the world that surpasses that of classic video generators.

The any-to-any architecture: how it works under the hood

The term "any-to-any" is often overused in AI. At Omni, it takes on a precise meaning: the model shares a common representation space for text, image, audio, and video.

A single encoder, a single decoder

Unlike approaches that stack specialized models (a CLIP for image, a Whisper for audio, an LLM for text), Omni uses a unified architecture. Any input is tokenized into the same latent space. The decoder then directly generates a video sequence with its audio track.

According to Storyboard18, this approach allows the model to maintain a temporal consistency that cascaded models struggle to achieve. The sound is not "added after" — it is generated simultaneously with the images, ensuring perfect synchronization.

Omni Flash: the lightweight model, not the weak model

The first public model, Omni Flash, is deliberately optimized for speed. Wavespeed reports that it generates 10-second clips with a latency compatible with interactive use. This is a strategic choice: Google prioritizes immediate accessibility over maximum quality for the launch.

Heavier Omni models, likely intended for later API access, should offer higher resolutions and durations. WinBuzzer confirms that broad API access is planned in a second phase, with paying users of Google apps being given priority.

Conversational editing: the real game-changer

Omni's most underestimated feature is conversational editing. You generate a video, then ask the model to modify it in natural language. No parameter tweaking, no mask painting, no keyframes.

A radically different workflow

With current models like Veo 3.1 or Seedance 2.0, if the video doesn't suit you, you modify your prompt and rerun a complete generation. It's iterative, costly in tokens, and frustrating when a single detail is the problem.

With Omni, you can say: "Replace the background with a beach at sunset" or "Make the character younger" or "Speed up the movement from the 5th second." The model modifies the existing video without regenerating everything. It's comparable to "inpainting" in images, but applied to video with complete semantic understanding.

According to Decrypt, Google positions this capability as central to the Omni experience. The idea: video becomes a living material that you sculpt through conversation.

Current limits of editing

We have to be honest: Omni Flash is in its early days. Complex modifications involving changes in physics (gravity, interactions between objects) or major structural changes remain tricky. Conversational editing shines on style, color, timing, and composition adjustments.

Omni vs the competition: Sora 2, Veo 3.1, Seedance 2.0

The video generation market in 2026 is a battleground between a few key players. Omni arrives with a specific positioning that deserves an honest analysis.

Comparison table of dominant video models (May 2026)

Where Omni wins

Input multimodality is unmatched. No competitor currently allows you to provide a photo, a musical excerpt, and a voice instruction to produce a video. Conversational editing is also a major differentiating advantage.

Ecosystem integration is the other strong point. Being available on YouTube Shorts from day one gives Omni a distribution surface that Sora 2 or Seedance 2.0 cannot match.

Where Omni loses (for now)

The raw quality of Omni Flash's 10 seconds probably does not beat Seedance 2.0 in 720p on visual fidelity benchmarks. And Veo 3.1 Audio 1080p likely remains superior in resolution and rendering quality for final outputs. Omni Flash is a speed-functionality trade-off, not a pure quality monster.

For a broader analysis of models, see our Claude, GPT, Gemini, Llama comparison: which model to choose in 2026?.

Concrete use cases: who gains what with Omni

YouTube Shorts and TikTok Creators

This is the most obvious and immediate use case. A creator can take a picture of their cat, add a text description, and get a 10-second clip with movement and sound. Directly in YouTube Shorts.

The time saving is considerable. What used to take 2-3 hours (filming, editing, sound design) can now take 5 minutes. LoraAI notes that YouTube Shorts is one of Omni Flash's three launch channels, which is no coincidence: Google wants to inject generative AI directly into its monetization pipeline.

Marketing and advertising

Marketing teams can prototype video visuals in a few conversational iterations. An image brief + voiceover + background music → animated video. Real-time modification with the client: "Change the product color", "Make the movement more dynamic".

For AI tools dedicated to marketing, check out our page on outils IA pour le marketing.

Social media and branded content

Agencies can generate variations of the same video for different networks: vertical format for TikTok, square for Instagram, with pacing adjustments requested in natural language. Our guide to outils IA pour les réseaux sociaux details other complementary solutions.

Prototyping for pro video production

Directors and studios can use Omni as a previsualization tool. Animated storyboard in a few minutes, testing of camera angles, exploration of soundscapes. The quality of Omni Flash is not sufficient for a final release, but the conversational editing workflow is perfect for ideation.

Integration into the Google ecosystem: Gemini, Flow, YouTube Shorts

Google is not launching an isolated model. Omni is embedded in three products simultaneously, and this integration strategy is probably more important than the model itself.

In the Gemini app

The most accessible interface. You chat with Gemini, you upload media, and the model generates an Omni Flash video in the conversation. Conversational editing is native: you continue chatting to modify the result. Available for Gemini Advanced subscribers.

To compare Gemini with other assistants, see our article Google Gemini vs ChatGPT vs Claude: which one for which use case?.

In Google Flow

Google Flow is Google's multimedia creation tool, focused on workflows. Omni is integrated there as a node in a larger pipeline: image generation with Gemini 3 Pro Image Preview, assembly with Omni, editing in Flow. This is where professionals will find the most value.

In YouTube Shorts

The most strategic integration. Any YouTube Shorts user can generate a video clip with Omni directly from the creation interface. Wavespeed confirms that access is free on YouTube Shorts, with usage limits. This is direct pressure on TikTok and its integrated AI tools.

The impact on content creation: revolution or evolution?

Omni isn't going to replace video creators tomorrow morning. But it accelerates an already heavy trend: the devaluation of technical "doing" in favor of creative "thinking."

What fundamentally changes

The barrier to entry for video creation is collapsing. Not because a model generates a perfect video — Omni Flash is far from that — but because the process becomes conversational. You no longer need to master Premiere Pro, After Effects, or animation principles. You need to have a clear idea and know how to communicate it.

Conversational editing also changes the creator-tool relationship. It's no longer a software with a complex interface; it's a collaborator that understands your instructions in natural language. Mashable describes Omni as a model capable of "creating anything", and while hyperbole is par for the course during Google announcements, the direction is clear.

What doesn't change (yet)

Quality. Omni Flash produces 10-second clips. That's enough for Shorts, not for long-form or professional content. Consistency over longer durations, managing recurring characters, complex transitions — all of this remains the domain of traditional tools or heavier models not yet public.

Originality. A model trained on existing data reproduces patterns. The videos generated by Omni will have a recognizable "AI look," just like all current generations. Human creativity remains essential to inject surprise and intention.

Strategic positioning: why Google is launching Omni now

Strategic reading is just as important as technical reading. Omni is not just a simple addition to the Google product line. It is a positioning maneuver.

Protecting the ecosystem

Seedance 2.0, ranked number one among video models by WaveSpeed, represents a threat to Google. If creators massively adopt external tools to generate content that ends up on YouTube, Google loses control of the value chain. Omni re-internalizes this step.

Differentiating from OpenAI

OpenAI's Sora 2 is powerful but remains a classic text-to-video model. By launching Omni as an any-to-any model with conversational editing, Google creates a distinct category in the minds of users. It is no longer about "generating a video", it is about "creating with a world model".

Fueling Gemini Advanced revenue

Omni Flash is a massive conversion argument for the Gemini Advanced subscription. When a model with this much media coverage is only available to paying users in the Gemini app, it creates leverage for sign-ups. WinBuzzer clearly details the paid-first rollout.

For free alternatives, our page of the best free LLMs lists the options accessible without a subscription.

What this announcement means for the future of multimodal models

Gemini Omni is not just a product, it's a strong signal of the direction generative AI is taking.

The end of single-modality models

The future does not belong to specialists. Models that only handle one type of input or output will gradually disappear. The trend is toward model families like Omni, which cover the entire multimodal spectrum from a single architecture. Separate rankings (image on one side, video on the other, audio apart) will become obsolete.

Editing as a first-class feature

For years, AI generation has focused on creating from scratch. Omni marks the shift to editing as a top-tier skill. This is a change as significant as the shift from plain text to rich formatting in LLMs.

Video as an interface

If Omni delivers on its promises, video could become a medium for communicating with AI on par with text today. You send a video of a problem, the model sends you back a video of the solution. You chat via video with an AI agent. It's speculative, but the direction is set.

To follow the evolution of the landscape, our article on new recent AI tools is continuously updated.

❌ Common mistakes

Mistake 1: Confusing Omni and Veo 3.1

These are two distinct model families. Veo 3.1 remains Google's high-quality video model, geared toward professional output. Omni is the any-to-any model focused on flexibility and editing. They coexist and serve different use cases. Failing to make this distinction means missing Google's positioning.

Mistake 2: Expecting cinematic quality with Omni Flash

Omni Flash is a fast and lightweight model designed for interactivity. The 10 seconds generated are suited for short-form content and prototyping. For high quality, Veo 3.1 Audio 1080p or external models remain more relevant. Judging Omni Flash on final quality criteria is like evaluating a sketch with the criteria of a finished painting.

Mistake 3: Ignoring the ecosystem aspect

Omni is not a model you consume via an isolated API (not yet, at least). Its value comes from its integration into the Gemini app, Google Flow, and YouTube Shorts. Isolating it from this ecosystem to compare it to an API-only model like Seedance 2.0 makes for a biased comparison.

Mistake 4: Believing that conversational editing replaces traditional editing

Omni's conversational editing allows for rapid semantic modifications. It does not replace narrative editing, creative transitions, fine sound design, or color grading. It is an ideation and prototyping tool, not a post-production suite.

❓ Frequently Asked Questions

Is Gemini Omni free?

Omni Flash is free on YouTube Shorts with usage limits. In the Gemini app and Google Flow, it requires a paid subscription (Gemini Advanced / Google One AI Premium). API access is not yet open to the general public.

What is the maximum duration of a video generated by Omni Flash?

Omni Flash generates 10-second clips. Future models in the Omni family are expected to support longer durations, but no date has been announced.

Does Omni Flash generate audio?

Yes, this is one of its key features. The audio is generated natively and synchronized with the video, without requiring a separate audio model.

Does Omni replace Veo 3.1?

No. Veo 3.1 remains Google's high-quality video model, particularly in 1080p with audio. Omni is complementary: it is more flexible with inputs and offers conversational editing, but Veo remains superior in pure rendering quality.

Can Omni be used via API?

Not at launch. WinBuzzer indicates that broad API access is planned for a later stage, with Google apps being the priority. For AI APIs available now, see our Free AI APIs page.

Is Omni better than Seedance 2.0?

In terms of raw video generation quality, Seedance 2.0 likely remains superior according to benchmarks. But Omni offers features that Seedance does not have: varied multimodal inputs, native audio, and conversational editing. The "best" depends on your use case.

✅ Conclusion

Gemini Omni doesn't do everything better than the competition, but it does something different: turning video into conversational and multimodal material. Natural language editing, integration into YouTube Shorts, and natively synchronized audio create a unique value proposition, even if Omni Flash isn't the most impressive model on the market. To follow all the announcements from this edition, check out our full coverage of Google I/O 2026: Gemini 4.0, Omni, Android XR and Aluminium OS.