← Back to workshop
Branch module
Google AI Studio Deep Dive
0%
✦ 0 XP 🌱 Curious
🌿 Optional branch — go deeper when you're ready
Branch Module · Google AI Studio

Gemini's free lab.
Use it properly.

AI Studio is Google's free platform for experimenting with Gemini — their most capable model family. It's where you prototype, test prompts, try multimodal inputs, and get an API key. Module 5 introduced it. This branch goes all the way in.

🔬 What AI Studio actually is

Google AI Studio (aistudio.google.com) is a free browser-based environment for working with Gemini models. It's not a product — it's a laboratory. You're talking directly to the model, with full control over the system prompt, temperature, context, and which model version you use.

It's different from Gemini.google.com (Google's consumer chatbot) in the same way a kitchen is different from a restaurant. Gemini.google.com serves you a meal. AI Studio puts you behind the stove.

Why this matters: Most AI tools put a layer of product between you and the model. AI Studio removes that layer. You can see exactly what's being sent to the model, tune the parameters, and get an API key to use the same model in your own tools — all for free.

The main model to know right now is Gemini 2.0 Flash: fast, free (with generous limits), multimodal (images, audio, video, PDFs), and genuinely capable. It's the right model for most things you'll build here.

🗺️ The interface — five things worth knowing

1
System instructions
The "System instructions" field (top of a new chat) is where you tell the model who it is and what its job is. This is different from your first message — it's a persistent instruction that shapes all responses. If you're building a tool that always answers in a specific format, this is where you define that.
2
Temperature
Controls how creative or deterministic the output is. 0 = very consistent, almost always the same response. 1+ = more varied, creative, sometimes surprising. For factual tasks: keep it low (0–0.3). For creative writing or brainstorming: raise it (0.7–1.2).
3
Model selector
Top-right dropdown. Gemini 2.0 Flash is your default. Gemini 1.5 Pro has a 2-million-token context window — useful if you're feeding it a very large document. Gemini 2.0 Flash Thinking is the reasoning model (slower but more thorough for complex problems).
4
Multimodal input
The paperclip icon in the chat lets you attach images, audio files, video clips, and PDFs. Gemini 2.0 Flash reads them natively — no conversion needed. You can ask questions about a photo, summarise a PDF, or transcribe an audio recording.
5
Get code
After a successful prompt, click "Get code" (top right of the response). AI Studio generates the Python, JavaScript, or REST equivalent of exactly what you just did — ready to paste into a project. This is how you turn a working prompt into a deployable tool.

🖼️ Multimodal — what you can actually do

Gemini's multimodal capabilities are genuinely useful beyond novelty. Here are the practical use cases:

Privacy note: What you upload to AI Studio goes to Google's servers. Don't upload sensitive personal information, client data, or anything you wouldn't want processed by a third party. For sensitive documents, use a local model (Ollama + Llama) or read the data privacy terms carefully.

🔑 Getting and using your API key

An API key lets you call Gemini from code — your own scripts, tools you build, integrations with other apps. Getting one takes 30 seconds.

1
Go to AI Studio → Get API key
aistudio.google.com → left sidebar → "Get API key" → "Create API key". Choose "Create API key in new project" if this is your first time.
2
Copy and store it safely
You'll only see the full key once. Copy it and store it in a password manager or a local text file. Never paste it into public code (GitHub, shared docs, forum posts). If you accidentally expose it, delete it immediately and create a new one.
3
Use it in a script
The simplest possible Python call to Gemini using your key:
Python — minimal Gemini call
pip install google-generativeai

import google.generativeai as genai
genai.configure(api_key="YOUR_API_KEY_HERE")
model = genai.GenerativeModel("gemini-2.0-flash")
response = model.generate_content("Summarise what vibe coding is in 3 sentences.")
print(response.text)

Free tier limits: Gemini 2.0 Flash on the free tier gives you 15 requests/minute and 1 million tokens/day. For most personal and small-business uses, you will never hit these limits. Paid tier (via Google Cloud) starts at $0.075 per 1M input tokens — very cheap for production use.

⚖️ AI Studio vs Claude vs ChatGPT — when to use which

Google AI Studio
Best for: building and experimenting
Free API access. Multimodal. Large context window. "Get code" turns your prompt into deployable code instantly. Best when you want to build something, not just chat.
Claude (Anthropic)
Best for: complex writing and reasoning
Most nuanced writing quality. Excellent at following complex instructions precisely. Best for: grant applications, long-form documents, careful reasoning tasks, anything where tone really matters.
ChatGPT (OpenAI)
Best for: image generation and breadth
DALL-E 3 integration. Widest third-party plugin ecosystem. Most familiar to non-technical users. Good for mixed tasks where you switch between text, images, and browsing.

Practical rule: Use AI Studio when you're building or need an API. Use Claude when the writing quality matters and you need it to follow precise instructions. Use ChatGPT when you need images alongside text, or when you're helping someone less technical who already has an account.

🎙️ NotebookLM — Google's secret weapon

NotebookLM (notebooklm.google.com) is a separate Google product built on Gemini, but it deserves a section here because it's one of the most immediately useful AI tools for businesses — and most people haven't heard of it.

The core idea: you upload your documents (PDFs, Google Docs, web pages, YouTube transcripts), and NotebookLM becomes an expert on those specific documents. It won't hallucinate facts from outside your sources — it stays grounded in what you gave it. Then it can do something remarkable: turn your documents into a podcast.

The Audio Overview feature: Give NotebookLM any set of documents and click "Generate Audio Overview." Two AI hosts will have a natural, conversational discussion about the key ideas — complete with analogies, back-and-forth, and summary. It sounds eerily real. A 40-page PDF becomes a 12-minute commute listen.

Real NZ business use cases:

Why it's better than just asking ChatGPT: NotebookLM cites sources — every answer links back to the specific passage in your document. You can verify it's not making things up. For legal, financial, or compliance documents, this is the only responsible way to use AI as a research tool.

TRY THIS
Turn one of your business documents into a podcast

Go to notebooklm.google.com (free, needs a Google account). Upload one document — a proposal, your website, a services PDF. Then:

  • Ask it a question in the chat — "What are the key services offered?" or "What should a new client know?"
  • Click Audio Overview → Generate. Wait 2–3 minutes.
  • Listen to the result. Notice the conversational tone.

For B2B use: NotebookLM is particularly powerful when you have a specific knowledge base to ground it in. A law firm's FAQ, a hospitality group's menu and policies, a school's parent handbook — the tool shines when the source material is defined.

👇 Your exercise

DO THIS NOW
Build a working AI Studio tool in 15 minutes

Pick one of these real use cases. Open AI Studio, write the system instruction, and test it.

  • Email reply assistant — System instruction: "You are a professional email assistant. When given an email, write a concise, warm, professional reply in NZ English. Don't start with 'I hope this finds you well.'" Then paste in a real email you need to reply to.
  • Menu description writer — System instruction: "You write enticing, honest menu descriptions. Under 30 words each. No food clichés (no 'succulent', 'artisan', 'decadent'). Warm and specific." Then give it your dishes.
  • Document Q&A — Upload a PDF (your lease, a council document, a contract). Ask: "What are the three most important things I need to know in this document? Are there any deadlines or action items?"

Once it works: click "Get code" and save the Python snippet. That's the start of a real tool.

Branch complete 🔬

AI Studio is the clearest path from "I tried the chatbot" to "I built something." The API key is the unlock — once you have it, you can connect AI to anything.