Claw field notebook
● last updated 2026-05-15 edit on GitHub colophon
Google / AI Studio / STUD.1 · 5 min read

Google AI Studio, plainly

Google AI Studio (aistudio.google.com) — the web playground for the Gemini API. Where you generate API keys, test prompts across mixed inputs (text, image, PDF, audio, video), prototype Live API voice/video, export code in five languages, and save prompts to your Drive. What it is, what it isn't, and the deprecated fine-tuning gap.

The thirty-second version#

Google AI Studio is the web playground for the Gemini API. Type a prompt, swap models, paste images / PDFs / audio / video, get a response, then click “Get Code” to export a working snippet in Python, Node, Go, Java, or REST.

Three things to know:

  1. It’s where you create your API key. The simplest path to using the Gemini API goes through aistudio.google.com/app/apikey. New users get a default Google Cloud project and a key in two clicks.
  2. It’s also a Live API testbed. The Stream Realtime tab gives you a working voice + video conversation with Gemini before you wire it into your own code. Helpful for spec’ing UX before commitment.
  3. No fine-tuning anymore. As of May 2025, the public Gemini API and AI Studio no longer support fine-tuning — Gemini 1.5 Flash-001 (the last tunable model) was retired. Any tutorial telling you to fine-tune in AI Studio is stale. Fine-tuning now lives in Vertex AI / Gemini Enterprise Agent Platform.

What you can do in AI Studio#

SurfaceWhat it is
Chat (Sandbox mode)Standard prompt-and-response. Upload images, PDFs, audio, video. Mixed-modality inputs in one box.
Stream RealtimeLive API testbed — bidirectional voice + video conversation with Gemini. Test interruptions (“barge-in”), tool calls, search grounding.
Build (apps surface)Prototype small Gemini-backed apps. The exact shape evolves; some sub-pages 404 as of mid-2026, suggesting it’s actively being rebuilt.
Image generationTest image models (gemini-2.5-flash-image-preview aka “Nano Banana”)
Video generationTest Veo models (Veo 3.1 lineup)
Music generationTest Lyria models
API keysCreate / restrict / delete keys at aistudio.google.com/app/apikey
Rate limits dashboardSee your account-specific RPM / TPM / RPD per model

Get Code — the headline feature#

Build your prompt, get it working, then click “Get Code”. AI Studio drops a working snippet for:

  • Python (uses the google-genai SDK)
  • JavaScript / Node.js (@google/genai)
  • Go (google.golang.org/genai)
  • Java (Maven artefact google-genai)
  • REST (raw curl)

The exported code uses the new unified SDK — it won’t generate the deprecated google-generativeai shape. That alone is a useful safety net if you’ve been pasting from older tutorials.

API keys — the bits that matter#

  • Create at aistudio.google.com/app/apikey
  • Tied to a Google Cloud project. AI Studio creates a default project for new users so you don’t have to think about it.
  • Display limits in AI Studio: up to 100 API keys and 50 projects are shown
  • Max 10 projects can be created at a time from the AI Studio Projects page (not a hard total cap — additional projects can be managed via the Cloud Console)
  • Google has flagged plans to enforce API-key restrictions during 2026 — unrestricted keys will be blocked. You’ll need to restrict each key to “Gemini API only” via AI Studio or the Cloud Console. Restrict your keys now to avoid surprises; check the API key docs for the current enforcement timeline.

Free tier — what’s included, what’s not#

The free tier on Gemini API direct (which is what AI Studio uses):

✅ Free of charge:

  • gemini-2.5-flash (text + image + video + audio I/O)
  • gemini-2.5-flash-lite
  • gemini-2.5-flash-live-preview (Live API)
  • gemini-2.5-flash-tts-preview
  • gemini-2.5-flash-image-preview (“Nano Banana”)
  • gemini-3-flash-preview
  • gemini-3.1-flash-lite (stable)
  • gemini-3.1-flash-tts-preview

❌ Paid tier required:

  • gemini-2.5-pro and gemini-3.1-pro-preview
  • gemini-3.1-flash-image-preview
  • veo-3.1 and veo-3.1-lite (video generation)

Specific RPM / TPM / RPD numbers aren’t published in the docs — Google directs you to aistudio.google.com/rate-limit where your live, account-specific limits are shown. They depend on the model and your usage tier.

Free tier privacy: when on the free tier, data may be used to improve Google products. On the paid tier, data is not used for training. Worth understanding before you paste sensitive content into the playground.

Saved prompts and sharing#

Prompts are saved into a Google Drive folder called AI_Studio in your account. From there:

  • Share a prompt with someone (link)
  • Download to migrate to Vertex AI Studio
  • Continue work on a different machine

Stream Realtime — the Live API testbed#

The Live API (covered in §GAPI.4 Patterns) is a WebSocket-based real-time voice + video shape. You can test it without writing any code:

  1. Open AI Studio → Stream Realtime tab
  2. Pick a Live API model (gemini-2.5-flash-live-preview is a good starting point)
  3. Click the mic, talk to Gemini
  4. Try interrupting mid-response (barge-in works)
  5. Try function calling — Live API supports it
  6. Try Google Search grounding — also supported

Once you’ve spec’d the UX, click “Get Code” to export a working starting point.

What AI Studio is not#

  • Not a production runtime — it’s a playground. Build your code with the exported SDK snippet; don’t rely on AI Studio sessions persisting indefinitely.
  • Not Vertex AI Studio (different product — see §STUD.2 vs API)
  • Not where you fine-tune Gemini anymore (deprecated May 2025)
  • Not where enterprise auth lives — for service accounts, ADC, Workload Identity, Agent Identity, you want Vertex AI Agents
  • Not free of data-use considerations — free tier data may be used to improve Google products

Honest take#

AI Studio’s strongest documented features are the small ones:

  • “Get Code” exports a working snippet using the current SDK (Google describes it as the “fastest path from prompt to production”)
  • The Realtime streaming tab is the documented way to feel out the Live API before committing to it
  • Default project + key creation in two clicks lowers the bar for new developers meaningfully

The frustrations are also small but real:

  • The App Builder / Build surface is mid-rebuild — sub-pages return 404, the docs lag the UI
  • Free-tier rate limits aren’t published — you have to log in to your dashboard to know your limits, which is a documentation gap
  • No fine-tuning is a real loss vs the older AI Studio era and isn’t well surfaced — many old tutorials still describe it
  • The page-naming flux (“Nano Banana” for the image model, “Lyria 3” for music in the sidebar nav) is internal-branding bleed-through that confuses readers

For a quick Gemini start, AI Studio is the right first stop. Get a key, build a prompt, copy the code, run it locally, iterate.

What’s next#

  • §STUD.2 vs API — AI Studio (UI) vs Gemini API (the thing AI Studio talks to) vs Vertex AI Studio (different product)
  • §GAPI.2 Getting started — turn the AI Studio prompt into running code
  • §GAPI.4 Patterns — Live API, structured output, context caching
  • §GEM.1 Gems — for “save a prompt as a reusable assistant inside the Gemini app” rather than “design and export code”

Sources