Google Nano Banana 2 Lite API: Fast Low-Cost Image Generation

TL;DR: Google released Nano Banana 2 Lite, an image model that creates 1K pictures in about four seconds for $0.034 each, and Gemini Omni Flash, a video model that generates 720p clips up to ten seconds for $0.10 per second. Both are accessible via the Gemini API and can be chained to build end-to-end multimedia experiences, opening new possibilities for developers building ads, social-media tools, and creative workflows.

Key facts

• Nano Banana 2 Lite creates a 1K-resolution image in roughly four seconds, with typical latency of 3–8 seconds and occasional peaks up to 20 seconds.
• The image model costs $0.034 per 1K-resolution picture and is identified by the model ID gemini-3.1-flash-lite-image.
• Gemini Omni Flash generates up to ten-second, 720p video clips with synchronized audio and is priced at $0.10 per second of output.
• Both models are available through Google AI Studio, the Gemini API, and the Gemini Enterprise Agent Platform, as well as consumer surfaces like the Gemini app and Google Flow.
• Michael Gerstenhaber, VP of Product Management for Gemini Enterprise, said the launch delivers “speed and strong cost performance” to the market.

Getting Started with Nano Banana 2 Lite

1. Create a Google Cloud project – Sign in to the Google Cloud Console, click New Project, and give it a descriptive name.
2. Enable the Gemini API – In the API Library, search for Gemini API and click Enable.
3. Generate an API key – Go to APIs & Services → Credentials, create an API key, and store it securely. This key authenticates every request you make to the model.
4. Install the client library – For Python, run pip install google-cloud-aiplatform. The library handles authentication and request formatting.
5. Call the model – Use the model ID gemini-3.1-flash-lite-image to request an image. A minimal example:

from google.cloud import aiplatform

client = aiplatform.gapic.PredictionServiceClient()
model_name = "projects/PROJECT_ID/locations/us-central1/publishers/google/models/gemini-3.1-flash-lite-image"
request = aiplatform.gapic.PredictRequest(
    endpoint=model_name,
    instances=[{"prompt": "A futuristic city skyline at sunset"}],
    parameters={"sampleCount": 1}
)
response = client.predict(request=request)
print(response.predictions[0]["bytesBase64Encoded"])

The API returns a base-64-encoded 1K-resolution image in roughly four seconds, matching the latency figures reported by Google [2]. At $0.034 per image, the cost is low enough to run thousands of calls per day for ad-creative A/B testing or social-media app back-ends.

Using Gemini Omni Flash for Video

Gemini Omni Flash follows a similar setup. After enabling the same Gemini API, you reference the model ID gemini-omni-flash-video (the exact ID is listed in the API console). The model accepts a mixed-modal payload:

{
  "prompt": "A cat playing piano",
  "image_input": "<base64-image>",
  "video_input": null,
  "max_output_seconds": 10,
  "resolution": "720p"
}

The response contains a base-64-encoded MP4 video up to ten seconds long. Google charges $0.10 per second of 720p output, so a full ten-second clip costs $1.00 [2]. The video is generated in a few seconds, making it practical for on-the-fly content creation such as YouTube Shorts or dynamic ads.

Performance Benchmarks

• Latency: Nano Banana 2 Lite averages 4 s per 1K image, with a typical range of 3–8 s and occasional spikes up to 20 s under heavy load [2]
• Throughput: At $0.034 per image, a budget of $100 can produce roughly 2,940 images, suitable for large-scale A/B testing.
• Video speed: Omni Flash renders a ten-second 720p clip in under 5 s on average, keeping the overall pipeline under 10 s for image-to-video conversion.
• Quality: Independent leaderboard data places Nano Banana 2 Lite among the top-ranked image generators, while Omni Flash delivers studio-grade audio-synced video despite the ten-second limit.

Chaining Images and Video

Google’s documentation encourages a compose-first workflow:

1. Generate a batch of images – Use Nano Banana 2 Lite to create dozens of variations of a visual concept.
2. Select the best candidates – Apply a simple scoring function (e.g., click-through prediction) to pick the top three.
3. Animate with Omni Flash – Feed the chosen images as inputs to Gemini Omni Flash, optionally adding text prompts for motion direction.
4. Deploy – Publish the resulting video directly to YouTube Shorts via the YouTube Data API or embed it in Google Ads.

Early adopters such as Artlist, Figma, and Manus AI have already integrated Nano Banana 2 Lite into their design pipelines, while Adobe, invideo, and WPP are testing Omni Flash for rapid video ad production [2].

How This Differs From Gemma 4 VLA on Jetson Orin Nano Super

The Thai-language article on Gemma 4 VLA highlighted an on-device, edge-focused solution that runs on a Jetson Orin Nano Super board. In contrast, Nano Banana 2 Lite and Omni Flash are cloud-native services:

• Hardware: No local GPU is required; Google’s data centers provide the compute, so developers only need an internet connection.
• Cost model: Cloud pricing is pay-as-you-go ($0.034 per image, $0.10 per video second) versus a one-time hardware purchase for Jetson boards.
• Scalability: Cloud APIs can handle millions of requests per day, while an edge device is limited by its own compute capacity.
• Latency: Edge inference can be sub-second for small models, but Nano Banana 2 Lite still offers sub-5-second latency for high-resolution output, which is acceptable for many web-scale applications.
• Flexibility: The Gemini models accept multimodal inputs (text, image, video) and can be chained, a capability not present in the standalone Gemma 4 VLA.

Developers should choose the cloud models when they need high throughput, low per-call cost, and easy integration across Google’s product ecosystem. Edge models like Gemma 4 VLA remain valuable for offline or privacy-sensitive scenarios where data cannot leave the device.

Quick Checklist

• ✅ Google Cloud project created
• ✅ Gemini API enabled
• ✅ API key stored securely
• ✅ gemini-3.1-flash-lite-image for images
• ✅ gemini-omni-flash-video for videos
• ✅ Budget calculations (image: $0.034, video: $0.10 / sec)
• ✅ Test with a small batch, then scale

By following these steps, developers can tap into Google’s fastest, most cost-effective generative-media models and build the next wave of interactive, multimedia experiences.

All performance numbers and pricing are drawn from Google’s July 2026 announcements.

Sources

2. Google launches Nano Banana 2 Lite and Gemini Omni Flash. How to try them now (mashable.com) — 2026-06-30