← Home
A VibeCoderLab cheat sheet
Hackathon build-day handout

AI Studiofor Dummies

aistudio.google.com · free · sign in with a Google account · no credit card

What this sheet is for

Keep it open beside the build window. When a word on the AI Studio screen makes no sense, look it up here — plain language, nothing else. The how-to lives in the deck; the why lives in the Vibe Coding guide. This is just the decoder.

The screen, word by word

Model
The AI brain doing the work. The default is Gemini 3.5 Flash. Flash is fast and free. Pro is smarter but slower and paid, so stay on Flash.
Prompt
The instruction you type. A clear one gets a working screen; a vague one gets a guess.
Build
The mode that turns your description into a real, working app.
Preview
The live app on the side of the screen. Tap around in it to test what you built.
AI Chips
Tap-in features you add to a prompt, like image generation or maps.
Annotation
Highlight a part of the rendered app and describe the fix in plain words.
Checkpoint
A saved point the agent pauses at before a big change, so you can approve or roll back.
Secrets
The panel where your API key is kept as a server-side secret, never in the app code.
Stream / Live
Talk to the model by voice or share your screen instead of typing.
Get Code
A button that exports your work as Python, Node, or cURL if you leave the studio.

The knobs & the engines

Token
A small chunk of text the model reads or writes. You only pay per token on the paid tier.
Context window
How much text the model holds at once. Flash holds about a million tokens — piles of notes.
System instructions
The role and rules you set up front, before the conversation starts.
Temperature
The creativity dial. Low for precise and repeatable, high for loose and inventive.
Structured output
Forces the answer into clean JSON your app can read, instead of free text.
Token limit
A cap on how long the reply can be.
Compare mode
Run one prompt across several models side by side to see which is best.
Grounding
Lets the model pull real facts — Search for the live web with citations, Maps for real places. One or the other, not both.
Function calling
Lets the model call an outside service so an agent can actually do things.
Code execution
Runs Python on the spot for sums and charts.
The models
Flash fast default · Pro deep reasoning · Nano Banana images · Veo video · Lyria music. All free to test; names change fast.
API key
The pass that lets an app talk to Gemini. It belongs to your project; making more keys does not add usage.
Firebase
The backend the studio wires up. Firestore remembers data; Authentication is the login.
Cloud Run
Where your app goes live. One click gives a shareable link.
Deploy
The act of putting your app online.
Looking for the how and the why?
One rule worth repeating: on the free tier your prompts may train Google's models — use fake demo data, never real personal details.