Gen AI Leader
Deep Dive
A top-down view: first the five layers that make up the generative AI ecosystem (infrastructure, models, platforms, agents, applications) with Google Cloud product examples at each one, then an in-depth look at Google's four proprietary foundation models — Gemini, Gemma, Imagen, and Veo.
Contents
The gen AI landscape can be thought of as a five-layer stack, each one built on top of the previous one. Understanding which layer each Google Cloud product lives in helps you pick the right tool for each business need.
1. Infrastructure
The hardware and compute that underpins everything else: GPUs and TPUs, orchestrated through Google Cloud's AI Hypercomputer, plus storage ( Cloud Storage) and high-capacity networking to train and run inference at scale.
2. Models
The foundation models themselves: Google's Gemini, Gemma, Imagen, and Veo, plus third-party and open source models available through Model Garden.
3. Platforms
The tools that let you build, customize, and deploy solutions on top of the models: Vertex AI Studio for prompt prototyping, Vertex AI Vector Search for semantic search, and Agent Platform for building agents.
4. Agents
Systems that use a foundation model together with function calling and external tools to reason, plan, and execute tasks autonomously, such as agents built with Agent Studio or accessible via Gemini Enterprise.
5. Applications
The end product aimed at the business user: already-packaged solutions like Conversational Agents or the Customer Engagement Suite, ready to solve a specific use case without the business team having to build on the lower layers.
Business implication
The further "down" the stack a team works (infrastructure, models), the more technical control it has but the more engineering effort it requires. The further "up" (applications), the faster business value is obtained but with less room for customization. The decision of which layer to build on is itself a strategic business decision.
Gemini
MultimodalGoogle's flagship foundation model, natively multimodal (text, image, audio, and video), available in different sizes depending on performance vs. cost/latency needs.
Gemma
Open modelFamily of lightweight, open-weight models, designed for local or edge deployment, or under the full control of the developer.
Imagen
Image generationGoogle's diffusion model specialized in generating and editing high-quality images from text.
Veo
Video generationGoogle's video generation model: produces high-quality clips from text and, optionally, a reference image.
The Gemini family is Google's flagship foundation model: natively multimodal (it processes and reasons over text, image, audio, and video in a single request) and available in different sizes depending on performance vs. cost/latency needs.
Use cases and strengths
Gemma is Google's family of lightweight, open-weight models, built using the same research as Gemini but designed to run in resource-constrained environments or under the full control of the developer.
Use cases and strengths
Imagen is Google's diffusion model specialized in generating and editing high-quality images from natural language descriptions.
Use cases and strengths
Veo is Google's video generation model, capable of producing high-quality video clips from text descriptions (and optionally a reference image).
Use cases and strengths
| Model | Primary modality | Use it when... |
|---|---|---|
| Gemini | Text, image, audio, video (multimodal) | You need complex reasoning, a large context, or to combine several modalities in a single request |
| Gemma | Text (primarily) | You need to run the model locally, at the edge, or with full control of the weights for budget or privacy reasons |
| Imagen | Image | The final deliverable is a static image: product, marketing, concept art, editing |
| Veo | Video | The final deliverable is moving content: clips, storyboards, promotional video |
To see where generative AI fits within Google Cloud's broader machine learning spectrum —from pretrained APIs to custom ML with Agent Platform (formerly Vertex AI)—, click on each level:
Did you get this topic?
Test what you just learned
A startup wants to run a language model directly on a portable medical device with limited connectivity, and needs full control over the model's weights to meet strict patient data privacy requirements. Which Google foundation model is best suited for this use case?
Put what you learned to the test with original Gen AI Leader exam questions.
Start Gen AI Leader mock examUnlimited mock exams with the Pro plan
Log in to track your progress.