AI is convergingand Google just hinted at its next big step

Google DeepMind CEO Demis Hassabis says the company will likely combine its flagship Gemini and Veo models over timea move that could reshape how we create and interact with video, and accelerate the rise of truly multimodal AI assistants. The remark, reported by TechCrunch, signals a clear direction: fewer, more capable models that natively understand and generate text, images, audio, and video.

If you are a founder, creator, or operator, heres what this shift means, why it matters, and how to prepare.

First, a quick refresher: Gemini vs. Veo today

Gemini: Googles multimodal foundation

Gemini is Googles family of multimodal models designed for reasoning across text, code, images, and audio. The Gemini 1.5 generation introduced a very long context window (hundreds of thousands to millions of tokens in research settings), enabling rich understanding across lengthy documents, recordings, or image sequences. Developers access Gemini across consumer products (e.g., Gemini in Google apps) and via the Google AI Studio/API.

Veo: Googles high-fidelity video generation model

Veo, unveiled at Google I/O 2024, is Googles most capable text-to-video model to date. It focuses on high-quality, longer-form shots with nuanced camera movement and adherence to stylistic prompts. Early access appears in creative tools like VideoFX, with more integrations planned across YouTube and Googles creative ecosystem.

In short: Gemini is the general-purpose multimodal brain; Veo is a specialist in generating cinematic video.

What Hassabis is signaling

According to TechCrunch, Hassabis expects Google to eventually combine Gemini and Veo. That doesnt necessarily mean a single model overnight; it suggests progressive unification of capabilities and interfaces so that one system can both reason about and generate video within the same workflow.

Google has already previewed this trajectory with initiatives like Project Astra, a real-time agent that perceives and reasons across modalities. Unifying Gemini and Veo would push that vision further: think assistants that can watch, understand, planand then produce high-quality video on the fly.

Why combine them? The strategic logic

• End-to-end multimodality: One system that can understand videos, keep long-term context, plan across steps, and generate new footage or edits without handing off to separate models.
• Lower integration complexity: Fewer moving parts for developers. Instead of stitching prompts across multiple APIs, a unified model can manage state, memory, and constraints.
• Better temporal reasoning: Video isnt just images; its dynamics over time. A combined model can reason about scene continuity, physics, and causality while generating or editing footage.
• Shared safety & controls: A single policy and moderation layer for both understanding and generation can strengthen guardrails around sensitive content, IP risks, and factuality.
• Efficiency at scale: Training and serving one frontier model family is costly, but consolidation can reduce duplication in infrastructure and improve latency through tighter optimization.

What could a unified Gemini+Veo unlock?

• AI video directors: Agents that interpret a product brief, generate moodboards, draft shots, iterate with your feedback, and deliver polished cuts with captions and voiceover.
• Live video tutors & coaches: Systems that watch your demo or gameplay in real time, explain whats happening, and produce tailored highlight reels with on-screen callouts.
• Enterprise knowledge to storytelling: Feed long meetings, documents, and dashboards into an assistant that outputs executive-ready explainer videos aligned to brand style.
• Search & understanding across footage: One agent that can watch hours of video, answer questions with citations, and synthesize summariesthen create a new training clip from what it learned.
• Creative collaboration: Think co-writing scripts with Gemini, then asking the same agent to block scenes and render shots in Veos stylewithout switching tools.

How soon could this happen?

Eventually implies a phased path rather than an immediate merge. Expect steps like:

• Tighter API orchestration: Seamless handoffs so developers call Gemini and Veo via a single workflow and shared safety settings.
• Shared memory and context: Long-context understanding that carries across analysis and generation tasks without re-prompting.
• Model family convergence: Over time, research directions for video understanding and video generation increasingly share architecture, training data strategies, and safety evaluations.

Google has already trended this way: multimodal reasoning in Gemini 1.5, real-time agents with Project Astra, and dedicated video generation in Veo. The next logical step is making these capabilities feel like one coherent system to the user.

Challenges Google must solve

• Compute & efficiency: Training and serving unified, frontier-scale models require massive compute and finely tuned inference paths to keep latency usable for interactive video workflows.
• Temporal consistency & realism: Long shots and edits must maintain character identity, motion coherence, and physical plausibility across scenes.
• Evaluation & safety: Robust benchmarks for video factuality, bias, and content safety are still maturing; unified systems need end-to-end auditing and watermarking.
• Copyright & provenance: As video gen scales, provenance signals and content labeling become critical for platforms and rights holders.
• Tooling & guardrails: Developers need APIs that expose power and controlstyle presets, brand safety, and editing constraints.

What this means for founders and teams

You dont have to wait for a full merge to capture value. Heres how to prepare now:

• Design for multimodal in/out: Organize your data and UX for text, image, audio, and video inputs and outputs. Build evaluation sets that reflect real tasks.
• Stay model-agnostic: Use an abstraction layer so you can switch between or combine Gemini, Veo, and other providers as capabilities evolve.
• Prototype workflows today: Chain Gemini for planning and Veo for generation via Googles APIs or tools like AI Studio; add human-in-the-loop for quality and safety.
• Invest in measurement: Track latency, quality (frame consistency, adherence to prompts), safety triggers, and production costs per deliverable.
• Mind IP & disclosure: Implement content provenance (e.g., C2PA where available), watermarking, and clear disclosures when outputs are AI-generated.

The competitive stakes

Google isnt alone in pursuing video-native, multimodal AI:

• OpenAI Sora: A text-to-video model that can produce realistic, longer scenes from prompts; early demos show strong physical and spatial reasoning. OpenAI
• Runway Gen-3: A production-focused video model emphasizing control, character coherence, and tool integrations. Runway
• Googles own stack: Alongside Veo, Google announced Imagen 3 for photorealistic images and continues to extend Gemini across search, workspace, and Android.

Unifying Gemini and Veo would position Google strongly in agentic, end-to-end video creationbut the race is active, and rapid advances across the field are likely.

Bottom line

Hassabiss comment points to a future where you wont think in terms of separate models for understanding vs. generating video. Youll just brief an AI partner, and it will watch, plan, and createall within one coherent experience. For builders and brands, the smart play now is to design multimodal workflows, keep your architecture flexible, and double down on measurement and safety. The stack will keep changing. Your readiness shouldnt.

FAQs

Will Gemini and Veo literally become one model?

Not necessarily overnight. Expect progressive unification: shared APIs, memory, and safety layersand, over time, deeper model-family convergence.

Can I use Gemini and Veo together today?

Yes. You can orchestrate Gemini for planning and Veo for generation via Googles developer tools. Expect tighter integrations as Google rolls out updates.

How does this compare to OpenAIs approach?

OpenAI is also pushing toward video-native, multimodal AI with Sora and agentic systems. Googles differentiation is long-context reasoning plus tight integration with its ecosystem (Search, YouTube, Android).

What about safety and copyright?

All major providers are investing in provenance, watermarking, and policy guardrails. As capabilities scale, expect stricter platform policies and clearer labeling.

What should startups do now?

Build model-agnostic pipelines, collect multimodal evaluation data, and ship human-in-the-loop workflows. Focus on specific use cases where video drives measurable outcomes.

Sources

1. TechCrunch: DeepMind CEO Demis Hassabis says Google will eventually combine its Gemini and Veo AI models
2. Google Blog: Introducing Veo, our most capable video generation model
3. Google Blog: Gemini 1.5our latest multimodal models with long context
4. Google DeepMind: Project Astra, towards real-time multimodal agents
5. The Verge: Google unveils Veo and Imagen 3 at I/O 2024
6. OpenAI: Introducing Sora
7. Runway Research: Gen-3 Alpha