A Gemini API key from Google AI Studio needs no billing account, which is part of why a fresh WordPress install already points at Gemini 3.1 Flash-Lite. The free plugin registers two Gemini models. Pro adds a third.
When to pick Google
Pick Google for bulk runs and for trying the plugin before committing budget. Gemini 3.1 Flash-Lite has the lowest blended price of any model the plugin registers: $0.56 per million tokens, against $2.00 for Claude 4.5 Haiku and $0.70 for GPT-4.1 mini. The trade-off is the preview-model limits described below, which bite hardest on exactly the bulk runs you picked Google for.
How to get a Gemini API key
Open Google AI Studio
Go to aistudio.google.com/app/apikey and sign in with any Google account — personal or Workspace both work. This is the same link the plugin puts on the API Keys page.
Create a new API key
Click Create API key. If you have several Google Cloud projects, pick one or accept the default. The key is generated immediately — copy it, because Google shows it once.
Paste it into the plugin
In WP Admin, go to Image SEO → API Keys, paste the key into the Google Gemini API card and click Save Gemini Key. The status badge flips to Configured, the key is stored encrypted, and the plugin verifies it with a live call.
Pick a Gemini model
A fresh install already runs gemini-3.1-flash-lite-preview, so you can stop here. To change it, open the AI Settings metabox on the plugin dashboard and pick from the AI Model dropdown — gemini-3.5-flash-lite on the free plugin, gemini-3.1-pro-preview with Pro. The provider follows from the model id, so there is no separate provider switch. Click Save Settings.
A key without billing is enough to start
You can create an AI Studio key without adding a billing account, and a fresh install already points at Gemini, so the plugin runs end to end on that key alone. The limits are real: the free quota is rate-limited, and a project with no billing can be refused on the preview models. Treat a no-billing setup as fine for a first pass, not for a 4,000-image bulk run.
What Save and Test Key check
Saving no longer checks what a key looks like. Since 1.0.6 the only rejections are structural: an empty value, fewer than 8 or more than 512 characters, or a value containing spaces or control characters. Anything else is stored encrypted.
Verification happens after the save, as a zero-cost request to the provider’s models list — https://generativelanguage.googleapis.com/v1beta/models?pageSize=1 for Google. The Test Key button on the provider card repeats that request on demand and reports one of three results.
Test Key results
| Setting | Description |
|---|---|
valid
|
The provider accepted the key. The message reads "Key verified with the provider." |
invalid
|
Google answered 400, 401 or 403. Its own error message is passed through unchanged, which is what tells you whether the key is wrong or the project is missing an entitlement. |
unverified
|
The check could not complete: a network failure, a timeout or any other status. The key stays saved and can be tested again later. This is never reported as invalid. |
Results are cached for 5 minutes, or 1 minute after a network failure, so repeated clicks do not hammer the provider. A key that fails the test is still stored — the result reports, it does not block.
A key prefix is orientation, not validation
Google keys have long started with AIza, which is useful for recognizing one in a password manager. The plugin checks nothing about the shape of a key, and Google is the reason: it changed its key format in 2026, and every format check written against the old one started rejecting valid keys. Only the live test tells you whether a key works.
Free-tier Gemini models
Google models in the free plugin
| Feature | Model id | Blended price / 1M tokens | Output cap, full run |
|---|---|---|---|
| Gemini 3.1 Flash-Lite Most cost-effective frontier model — the default in a fresh install, and a preview model. | gemini-3.1-flash-lite-preview | $0.56 | 2,048 tokens |
| Gemini 3.5 Flash-Lite Newer generation, cost-effective. Added in 1.0.6, about 1.5 times the price of 3.1 Flash-Lite. | gemini-3.5-flash-lite | $0.85 | 2,048 tokens |
Both sit in the Google group of the AI Model dropdown, 3.1 Flash-Lite first and 3.5 Flash-Lite under it. The dropdown reads the version number out of the model id and lists the newest model last within each provider group. A Gemini added later lands at the bottom of the group, not wherever it happened to register.
Preview models carry preview limits
Gemini 3.1 Flash-Lite and Gemini 3.1 Pro are preview models — both ids end in -preview. Preview endpoints run tighter rate limits than general-availability models, and a project without billing enabled can get a 403 back. That surfaces on a bulk run long before it surfaces on a single image. Gemini 3.5 Flash-Lite is not a preview model.
Installs that saved the older id gemini-3-flash-preview keep working: it is registered as an alias of 3.1 Flash-Lite, so the saved value resolves to the current model with no action from you.
Pro-only Gemini models Pro
Pro adds Gemini 3.1 Pro
With Image SEO Optimizer Pro active, Gemini 3.1 Pro joins the Google group in the AI Model dropdown. At $2.50 per million tokens it runs about 4.5 times the blended price of 3.1 Flash-Lite and about 3 times that of 3.5 Flash-Lite, with a far larger output cap — 8,192 tokens against 2,048. It is also a preview model, so the rate-limit warning above applies to it too.
Google model added by Pro
| Feature | Model id | Blended price / 1M tokens | Output cap, full run Recommended |
|---|---|---|---|
| Gemini 3.1 Pro State-of-the-art multimodal reasoning — the description the plugin shows in the dropdown. | gemini-3.1-pro-preview | $2.50 | 8,192 tokens |
Pro registers Gemini 3.1 Pro through the same hook a third-party plugin would use, and it does not change the default model — a Pro site still runs Gemini 3.1 Flash-Lite until you pick something else. Output varies between runs on every model here; none of them is deterministic.
These prices are per million tokens, not per image
The figures on this page are the blended per-million-token prices the plugin stores for each model and uses to estimate spend. Cost per image depends on image size, prompt length and how many fields you generate, so a per-image number would be a guess. The API Usage dashboard logs every call and shows what you actually spent. Checked against version 1.0.6 on 2026-08-12.
Next steps
Optimization Settings
Pick a model, batch size and request timeout for your chosen provider.
Prompts & Language
Tune the AI's voice with a prompt addon and pick from 133 output languages.
API Usage & Costs
Live dashboard of every AI call, per-provider breakdown, monthly spend.