When it comes to AI image generation and editing, finding the right balance between quality, speed, and cost can feel like searching for a needle in a haystack. I’ve spent countless hours testing various image models, and recently stumbled upon something that genuinely caught my attention: Gemini 2.5 Flash Image, affectionately nicknamed “Nano Banana” by the developer community.
What makes this particularly interesting isn’t just the model itself—it’s how OpenRouter has made it accessible through a familiar OpenAI-compatible API. This means you can integrate advanced image generation and editing capabilities into your existing workflows without rewriting your entire codebase.
Understanding Nano Banana: More Than Just Another Image Model
Gemini 2.5 Flash Image represents a significant leap forward in controllable image generation. Unlike traditional image generators that often struggle with precise edits or maintaining consistency across variations, this model excels at targeted modifications while preserving the essence of your original content.
The real breakthrough lies in its ability to handle multi-image fusion with remarkable character consistency. Whether you’re merging elements from different shots or maintaining brand identity across dozens of product variations, Nano Banana delivers results that feel intentional rather than algorithmic.
OpenRouter partnered with Google to launch this model, making it the first broadly accessible image model through their unified API gateway. This partnership opens doors for developers who want enterprise-grade image capabilities without juggling multiple API integrations or wrestling with vendor-specific implementations.
Technical Setup: Getting Started with OpenRouter
The beauty of this implementation lies in its simplicity. Since OpenRouter maintains OpenAI compatibility, you can use the official SDK with minimal configuration changes. Here’s what you need to know:
| Configuration Item | Details |
|---|---|
| Model Identifier | google/gemini-2.5-flash-image-preview |
| API Endpoint | /api/v1/chat/completions |
| Base URL | OpenRouter |
| Supported Input | Text and image modalities |
| Image Input Method | URL or base64 data URI through image_url content type |
| Output Format | Image data returned in chat message content |
| Optional Headers | HTTP-Referer and X-Title for leaderboard attribution |
Start by obtaining an API key from OpenRouter and configure your client to point at their base URL. The optional headers help attribute your traffic and can get your application featured on OpenRouter’s leaderboards, though they’re not required for core functionality.
Generating Fresh Images: A Practical Approach
Creating new images from scratch involves sending a chat completion request with both text and image modalities enabled. The model processes your text prompt and returns image data within the standard chat response structure.
Here’s a straightforward cURL example that demonstrates the core pattern:
curl https://openrouter.ai/api/v1/chat/completions \
-H "Authorization: Bearer $OPENROUTER_API_KEY" \
-H "Content-Type: application/json" \
-H "HTTP-Referer: https://example.com" \
-H "X-Title: Nano Banana Demo" \
-d '{
"model": "google/gemini-2.5-flash-image-preview",
"modalities": ["text", "image"],
"messages": [
{
"role": "user",
"content": [
{ "type": "text", "text": "Generate a photorealistic street scene at dusk with warm neon signage and light rain." }
]
}
]
}'
The response includes your generated image embedded in the message content. You’ll need to iterate through the choices array and extract the image item for saving or further processing.
Image Editing: Where Nano Banana Truly Shines
While generating images from scratch is useful, the real power emerges when you need precise, controlled edits to existing visuals. Nano Banana’s semantic understanding allows for remarkably nuanced modifications—think “add a subtle neon reflection” or “change the sign to soft cyan while keeping the subject’s face consistent.”
The editing workflow accepts input images through publicly accessible URLs or base64-encoded data URIs. You can even include multiple images in a single request, enabling sophisticated multi-image fusion workflows:
INPUT_IMAGE_URL="https://example.com/original.jpg"
curl https://openrouter.ai/api/v1/chat/completions \
-H "Authorization: Bearer $OPENROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "google/gemini-2.5-flash-image-preview",
"modalities": ["text", "image"],
"messages": [
{
"role": "user",
"content": [
{ "type": "image_url", "image_url": { "url": "'"$INPUT_IMAGE_URL"'" } },
{ "type": "text", "text": "Add a subtle neon reflection on the wet pavement, change the sign to soft cyan, and keep the subject's face consistent." }
]
}
]
}'
This approach feels natural because you’re essentially having a conversation about what you want changed, rather than fiddling with abstract parameters or learning model-specific syntax.
Integrating with Modern Development Stacks
For JavaScript developers working with Node.js applications, the OpenAI SDK integration is seamless:
import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.OPENROUTER_API_KEY,
baseURL: "https://openrouter.ai/api/v1",
});
const resp = await client.chat.completions.create({
model: "google/gemini-2.5-flash-image-preview",
modalities: ["text", "image"],
messages: [
{
role: "user",
content: [
{ type: "text", text: "Generate a cozy reading nook by a rainy window, warm tungsten lighting, muted teal palette." },
],
},
],
});
Python developers enjoy the same straightforward experience:
from openai import OpenAI
import os
client = OpenAI(
api_key=os.environ["OPENROUTER_API_KEY"],
base_url="https://openrouter.ai/api/v1",
)
resp = client.chat.completions.create(
model="google/gemini-2.5-flash-image-preview",
modalities=["text", "image"],
messages=[
{
"role": "user",
"content": [
{"type": "text", "text": "Create a modern product hero shot on a glossy black surface with soft rim lighting."}
],
}
],
)
Both implementations follow identical patterns for image editing—simply add image_url content parts before your text instructions.
Practical Insights and Common Pitfalls
Through real-world testing, I’ve encountered a few scenarios that trip up developers new to this workflow. If your generation returns only text without images, double-check that your modalities array explicitly includes “image” and verify the model supports image output in OpenRouter’s current model roster.
Prompt quality significantly impacts results. Vague directives like “make it better” produce inconsistent outputs, while specific instructions such as “soften the background by 30%, add rim lighting from the left, maintain the product’s original color temperature” yield reliable, reproducible results.
The model’s character consistency proves invaluable for high-volume editing workflows. Several teams have leveraged Nano Banana for generating hundreds of ad variations from a single source image, maintaining brand consistency while exploring different compositions and color treatments.
Understanding the Economics
Pricing follows OpenRouter’s unified model, which may differ from direct API access. Google’s official announcement positions Gemini 2.5 Flash Image at approximately $0.039 per image through their native API, though OpenRouter’s routing layer and pricing structure provides its own value proposition through unified access to hundreds of models.
For projects requiring diverse AI capabilities, consolidating through OpenRouter often simplifies billing and integration complexity, even if per-image costs vary slightly from direct provider access.
Why This Approach Matters
The convergence of large language model world knowledge with diffusion-quality image output creates genuinely new capabilities. Traditional image models struggle with semantic concepts like “make this feel more corporate” or “adjust the mood to match late autumn.” Nano Banana’s LLM foundation enables these conceptual transformations while maintaining technical precision.
OpenRouter’s compatibility layer means the same codebase that generates images can also handle text completion, embeddings, or any other model-based task. This architectural consistency reduces cognitive overhead and accelerates development cycles.
Transform Your Digital Vision into Reality
Whether you’re building the next generation of creative tools, automating visual content workflows, or exploring new ways to engage your audience, mastering modern image generation APIs opens remarkable possibilities.
At Tenten, we specialize in transforming complex technical capabilities like these into practical business solutions. Our team has deep experience integrating AI-powered image generation into everything from e-commerce platforms to marketing automation systems. We understand both the technical implementation and the business strategy needed to extract real value from these emerging technologies.
If you’re ready to explore how advanced image generation could elevate your digital presence or streamline your creative workflows, let’s talk. Book a meeting with our team and discover how we can help you harness these powerful capabilities for your unique business needs.