Nano Banana Pro API is a market-facing name that developers use for Google's Pro image route. In the official Google stack, the current mapping is Gemini 3 Pro Image with model ID `gemini-3-pro-image`. The practical decision is not only how to send a request. It is which contract owns the request: Google Standard, Google Batch/Flex, a provider gateway, or a cheaper Nano Banana model that makes Pro unnecessary.
Google Standard image output, checked on July 1, 2026, is `0.134\` for \`1K/2K\` and \`0.24` for `4K`. Google Batch/Flex is the first-party discount lane at `0.067\` and \`0.12`, but it changes delivery expectations. Public laozhang.ai docs checked the same day list `0.09/image\` and \`0.09/edit`; that is provider pricing, not official Google pricing, and it needs console proof before production.
Быстрое решение можно свести к такой таблице:
| Route | Когда использовать | Кто владеет ценой или proof | Stop rule |
|---|---|---|---|
| Google Standard | Нужен официальный real-time Gemini API. | Google pricing docs, checked July 1, 2026. | Не называйте provider price скидкой Google. |
| Google Batch/Flex | Можно ждать и нужен самый дешевый first-party Pro lane. | Google Batch/Flex pricing docs. | Не использовать как real-time Standard. |
| Provider gateway | Нужны billing flow, compatible endpoint, logs or access simplification. | Provider docs, console and call logs. | Проверить alias, price, failed-call billing, status and data terms. |
| Cheaper Nano Banana model | Pro fidelity is not required. | Google model family and live console rows. | Не платить Pro price for drafts or default workloads. |
These facts were checked on July 1, 2026 against Google image-generation, pricing, API-key, billing and rate-limit docs, plus public laozhang.ai Nano Banana Pro docs. Prices, aliases, quotas and provider billing can change. Recheck the official source or the provider console before launch.
Что сейчас означает Nano Banana Pro API
Nano Banana Pro API is not a separate official product name. It is the search and developer shorthand for the Pro image model. In implementation, the current official Google model is Gemini 3 Pro Image, and the model string is `gemini-3-pro-image`. Older `gemini-3-pro-image-preview` strings belong only in migration notes or provider compatibility notes unless current docs explicitly require them.
This distinction is operationally important. A working image request has a model owner, a contract owner and an operating owner. The model owner is the image model. The contract owner decides pricing, billing, support and failure rules. The operating owner is the project, key, quota, logs and status route that will explain errors. Many teams debug the wrong thing because these three owners are collapsed into one line of code.
For an early test, AI Studio plus a Gemini API key can be enough. For production, the same route needs a billing project, current key type, quota tier, retry policy, request logs and cost instrumentation. Code alone cannot prove that the path is safe to run at scale.
Сначала решите, действительно ли нужен Pro
Nano Banana Pro should be the override model, not the default for every image. Use `gemini-3-pro-image` when the job needs 4K output, complex composition, reliable typography, brand assets or final product imagery. Use a cheaper Nano Banana model when the job is draft generation, exploration, internal thumbnails, high-volume variants or simple product scenes.
| Model family | Current API model ID | Best fit | Decision rule |
|---|---|---|---|
| Nano Banana 2 Lite | `gemini-3.1-flash-lite-image` | Cheap drafts and fast iteration. | Start here when cost and speed matter most. |
| Nano Banana 2 | `gemini-3.1-flash-image` | Most default production images. | Let it fail the quality bar before moving to Pro. |
| Nano Banana Pro | `gemini-3-pro-image` | 4K, layout, text fidelity, high-value assets. | Use it when lower models visibly fail the job. |
| Legacy Nano Banana | `gemini-2.5-flash-image` | Existing integrations that have not migrated. | Keep only for compatibility and plan migration. |
The strongest cost control is often model substitution. A production pipeline can use cheaper models for ideation and candidate generation, then reserve Pro for final render or approval. That design is easier to explain to finance than sending all traffic to a premium model and later trying to hide the cost in gateway shopping.
Официальные цены Google: Standard отдельно, Batch/Flex отдельно
Google pricing should stay in its own table. Provider rows belong in a separate provider section, because a provider price is a different contract.
| Official Google lane | `1K/2K` image output | `4K` image output | Meaning |
|---|---|---|---|
| Standard | `$0.134` | `$0.24` | Official real-time Pro route. |
| Batch/Flex | `$0.067` | `$0.12` | Cheaper first-party route when async or flexible processing is acceptable. |
The key point is why Batch/Flex is cheaper. You are not finding a hidden coupon. You are choosing a different first-party processing lane. If the product needs immediate user-facing output, Standard is the safer route. If the workload is offline, queued, periodic or batch-generated, Batch/Flex is the first-party cost lane to evaluate.
Two boundaries should stay next to every estimate. First, the checked Google pricing page did not list a free image-output tier for `gemini-3-pro-image`. Second, grounding can add separate search-query charges. An estimate that only counts the image row can still understate the real bill.
Key, billing and quota are project boundaries
An AI Studio API key is only the authentication surface. It is not the full billing and quota plan. Current Google key docs say new AI Studio keys are auth keys; unrestricted standard keys have been rejected from June 19, 2026, and standard keys are rejected in September 2026. Old code examples that rely on older key behavior should not become the new production baseline.
| Check | Why it matters | Where to verify |
|---|---|---|
| Key type | Older unrestricted-key examples can fail. | AI Studio key page and Google API-key docs. |
| Billing project | Paid API usage belongs to the linked project, not to the code snippet. | Google Cloud billing and AI Studio billing views. |
| Quota owner | Rate limits are project-level, not per API key. More keys do not multiply quota. | Gemini API rate-limit docs and console quota pages. |
| Tier | Tier 1 needs active billing; Tier 2 needs $100 paid plus 3 days; Tier 3 needs $1,000 paid plus 30 days. | Gemini rate-limit docs checked July 1, 2026. |
| Live limit | RPD, RPM, TPM and image-specific limits can differ by model and project. | Console live limits before load testing. |
If you hit quota, key sprawl is the wrong fix. Check live project limits, add jittered backoff, batch non-urgent work, lower concurrency, move suitable jobs to Batch/Flex and request quota through the official route. Key rotation helps security. It does not create more project quota.
First official request with the Interactions API
Google's current image docs recommend the Interactions API for latest features and models. The examples below use `gemini-3-pro-image`. Keep `GEMINI_API_KEY` in an environment variable.
Python
```python from google import genai import base64
client = genai.Client()
interaction = client.interactions.create( model="gemini-3-pro-image", input="Create a premium product hero image for a black mechanical keyboard on a clean studio background.", response_format={ "type": "image", "aspect_ratio": "16:9", "image_size": "2K", }, )
if interaction.output_image: with open("nano-banana-pro-output.png", "wb") as f: f.write(base64.b64decode(interaction.output_image.data)) ```
JavaScript
```javascript import { GoogleGenAI } from "@google/genai"; import * as fs from "node:fs";
const ai = new GoogleGenAI({ apiKey: process.env.GEMINI_API_KEY });
const interaction = await ai.interactions.create({ model: "gemini-3-pro-image", input: "Create a premium product hero image for a black mechanical keyboard on a clean studio background.", response_format: { type: "image", aspect_ratio: "16:9", image_size: "2K", }, });
if (interaction.output_image) { fs.writeFileSync( "nano-banana-pro-output.png", Buffer.from(interaction.output_image.data, "base64"), ); } ```
REST
```bash
curl -s -X POST
"https://generativelanguage.googleapis.com/v1beta/interactions"
-H "x-goog-api-key: $GEMINI_API_KEY"
-H "Content-Type: application/json"
-d '{
"model": "gemini-3-pro-image",
"input": [
{
"type": "text",
"text": "Create a premium product hero image for a black mechanical keyboard on a clean studio background."
}
],
"response_format": {
"type": "image",
"aspect_ratio": "16:9",
"image_size": "2K"
}
}'
```
If an existing codebase still uses `generateContent`, migrate deliberately. Do not change the model, endpoint, provider, image size and response parser at the same time. For image sizes, keep the `K` uppercase in `1K`, `2K` and `4K`.
Когда имеет смысл gateway-провайдер вроде laozhang.ai
A provider gateway can be useful when the team wants OpenAI-compatible billing, a simpler payment path, gateway logs, multi-model routing or a fallback access route. laozhang.ai belongs only in that provider lane for API access, cost, quota and implementation routing.
The boundary is strict: provider pricing is not official Google pricing. Public laozhang.ai docs checked on July 1, 2026 list `0.09/image\` and \`0.09/edit`, and still expose provider alias language. Before production, the provider console, model list, request log and billable event rule must be the source of truth.
| Proof item | What to record | Pass condition |
|---|---|---|
| Endpoint and model alias | Base URL, request shape, provider model name, returned model name. | Alias maps clearly to the intended Pro model. |
| Console price and billing mode | Unit price, recharge or balance mode, generation and edit differences. | Finance accepts the console price. |
| Failed-call rule | Timeout, safety block, upstream error and partial output behavior. | Rule is documented or observable in logs. |
| Rate and status route | Rate-limit headers, status page, support route and incident history. | You know where to look before blaming Google. |
| Data boundary | Retention, region, content policy and support terms. | Terms fit the workload and customer promise. |
| Small logged test | Prompt, response, latency, cost and status code. | Logs match the claim before scale. |
The safe recommendation is conditional. Use laozhang.ai only when current console and logs satisfy your model, price, billing, data and failure requirements. Do not promise unlimited speed, no blocking, failed-call refunds or guaranteed availability without current proof.
Troubleshooting the first request
Keep the branch stable when the first request fails. Changing key, model, provider and payload at the same time destroys the diagnostic signal.
| Symptom | First check | Likely owner | Next move |
|---|---|---|---|
| `401` or key rejected | Key type, restrictions, project link, auth-key migration. | Credential setup. | Create or migrate the key in AI Studio and retest the same request. |
| `403` or permission wording | Project, billing, policy, file access, provider denial. | Project or policy owner. | Separate permission from safety and file-resource errors. |
| `429 RESOURCE_EXHAUSTED` | Project/model live limits, RPM, TPM, RPD and image limits. | Quota and traffic shape. | Back off, batch non-urgent work or request quota. |
| No image or text-only output | Response parts, modality, safety finish and prompt shape. | Request and parser. | Inspect output fields before calling it an outage. |
| `503/504` or overload | Same-path retest, provider status, Google status and timeout. | Capacity or gateway path. | Retry once with bounds, then switch route only with logs. |
| Bill higher than estimate | Resolution, grounding, retries, duplicate jobs and provider failed-call rule. | Cost instrumentation. | Add per-request cost logs before launch. |
Preserve a minimal reproduction: model ID, route, project, key type, image size, prompt, status code, request ID and billing evidence. Without that packet, the next move is guesswork.
Production checklist before launch
Before launch, assign each owner:
- Model owner: `gemini-3-pro-image` is necessary, or a cheaper model has failed the quality bar.
- Price owner: Google Standard, Google Batch/Flex or provider console price is recorded.
- Credential owner: Auth key, restrictions and rotation policy are clear.
- Quota owner: Live project/model limits and tier requirements are checked.
- Cost owner: Resolution, retries, grounding, failed calls and provider billable events are logged.
- Data owner: Google or provider terms allow the image inputs you send.
- Fallback owner: You know when to lower resolution, move to Batch/Flex, switch to Nano Banana 2 or change provider.
- Rollback owner: A cheaper model or disabled feature flag is ready before traffic rises.
The practical rule is simple: choose the contract owner, prove price and quota, then copy code. That order keeps Nano Banana Pro API from becoming an expensive debugging exercise.
Часто задаваемые вопросы
Какой model ID использовать?
Use `gemini-3-pro-image` for the current official Google Pro image model. Treat `gemini-3-pro-image-preview` as legacy or provider alias wording unless active docs require it.
Есть ли бесплатный Pro image API?
The Google pricing page checked on July 1, 2026 did not list a free image-output tier for `gemini-3-pro-image`. AI Studio can be free as a product surface; paid API usage is still billable.
Какие официальные цены Google?
Standard image output is `0.134\` for \`1K/2K\` and \`0.24` for `4K`. Batch/Flex lists `0.067\` and \`0.12`. Recheck before production.
laozhang.ai это official Google pricing?
No. It is a provider gateway. Its public `$0.09/image` row is provider pricing and must be verified against console and logs before production.
Больше API keys увеличивают quota?
No. Google rate limits are project-level, not per key. More keys can help separation and rotation, but they do not create more quota.
Когда выбирать Batch/Flex?
Choose Batch/Flex when you need Google's first-party contract and can accept async or flexible delivery. Do not use it for immediate interactive output.
Какие источники перепроверить позже?
Recheck Google image-generation, pricing, API-key, billing and rate-limit docs. For provider routes, recheck provider docs, console, model list and call logs.