Masterclass: Multi-Model AI Image Generation Routing — All 6 Gemini Models

01 // Quick Answer

How do you route across all 6 Gemini image generation models?

Quick Answer — Voice & AI Search Target

Route by purpose, not by price. All 6 models solve different problems, and the router picks the specialist for each job.

In April 2026, the Gemini API ships six image generation models spanning three use cases. For drafts and previews: Imagen 4 Fast at $0.02 per image. For high-volume economy: Nano Banana (gemini-2.5-flash-image) at $0.039 per image. For general heroes: Imagen 4 Standard at $0.04. For print-quality finals: Imagen 4 Ultra at $0.06. For high-throughput contextual generation: Gemini 3.1 Flash Image at $0.067 per 1K image. For character-consistent hero content: Nano Banana Pro (gemini-3-pro-image-preview) at $0.134 per 1K, $0.24 per 4K. AGI-CORE-Pro defaults 65 percent of traffic to Nano Banana, uses the Batch API for a 50 percent discount on non-interactive jobs, and chains Nano Banana into Nano Banana Pro for identity-locked outputs.

02 // Context

Three Families, Six SKUs, One Routing Decision

Google ships image generation through three distinct model lines on the Gemini API: Nano Banana (speed + context), Nano Banana Pro (quality + identity), and Imagen 4 (print quality + text rendering). Add Gemini 3.1 Flash Image for high-throughput contextual work and you have six specialists. Each wins a different job.

The image generation landscape in 2026 is not a single "best model" problem. It's a routing problem. Nano Banana wins on per-image price for contextual conversational workloads. Nano Banana Pro is the only model on the Gemini API with native reference-image support for identity preservation across a catalog. Imagen 4 ships three SKUs — Fast, Standard, Ultra — that lead the family on pure text rendering and print-quality output. Gemini 3.1 Flash Image sits between the two, offering Gemini 3's prompt understanding at higher throughput than Nano Banana Pro.

The question is no longer "which model do I use?" It's "which specialist wins each job bucket, and how do I dispatch among them without leaking cost?" A solo founder running 10,000 product images per month pays $1,340 on Nano Banana Pro alone, $400 on Imagen 4 Standard, $390 on Nano Banana, or $200 on Imagen 4 Fast — a 6.7x spread on pricing for the same volume. Route wrong, bleed runway.

This masterclass walks the 6-model matrix end to end: the purpose axes, the price ladder, the character-consistency chain, the Batch API savings, and the exact SDK code you need to ship a production router on the google-genai SDK.

03 // 6-Model Matrix

The Full Gemini Image Generation Family

Six specialists. One router. Each card below maps a model to its exact Gemini API rate card price, its best-fit job, and the constraint that triggers selection.

Key Takeaways — 6-Tier Selection Logic

How the six Gemini image models map to production workloads

Nano Banana is the default route — $0.039 per 1K image. Economy tier. 65 percent of all generations land here. Text input at $0.30 per 1M tokens.
Imagen 4 Fast is the draft tier — $0.02 per image, cheapest standalone model in the family. Preview thumbnails, A/B creative tests, rapid iteration.
Imagen 4 Standard is the general hero — $0.04 per image. Superior text rendering. Product shots, blog heroes, marketing images.
Imagen 4 Ultra is the print-quality final — $0.06 per image. Billboard-scale assets. Logo and typography work. Physical reproduction-grade output.
Gemini 3.1 Flash Image is the high-throughput contextual tier — $0.045 at 0.5K, $0.067 at 1K, $0.101 at 2K, $0.151 at 4K. Conversational image editing and contextually aware variations.
Nano Banana Pro is the character-lock tier — $0.134 per 1K/2K, $0.24 per 4K. Native reference image support (up to 3). The only model that preserves subject identity across a catalog.
Batch API delivers 50 percent off every model — Imagen 4 Fast drops to $0.01, Nano Banana drops to $0.0195, Nano Banana Pro drops to $0.067 per 1K. Use for non-interactive bulk jobs.
All 6 models are synchronous — no LRO polling, unlike Veo 3.1 video. Single generate_content call returns the image directly.

Nano Banana Pro

gemini-3-pro-image-preview

Character Lock

PurposeHero + identity lock

Reference ImagesUp to 3

Text RenderingExcellent

Sync / AsyncSynchronous

1K / 2K image$0.134

4K image$0.24

Text input$2.00 /1M

Ref image input$0.0011 /img

Premium tier. The only Gemini image model with native reference-image conditioning. Reserve for character-consistent product catalogs, hero content, and final render pipelines. Batch API: $0.067/1K.

Gemini 3.1 Flash Image

gemini-3.1-flash-image-preview

High-Throughput

PurposeContextual edits

Prompt ModelGemini 3.1

Sync / AsyncSynchronous

0.5K (512px)$0.045

1K (1024px)$0.067

2K (2048px)$0.101

4K (4096px)$0.151

Text input$0.50 /1M

High-throughput contextual generation with Gemini 3's prompt understanding. Best for conversational image editing and contextually-aware variations. Batch API halves all prices.

Nano Banana

gemini-2.5-flash-image

Economy

PurposeEconomy / default

Max Resolution1024 x 1024

Token Cost1290 tokens/image

Sync / AsyncSynchronous

Per image$0.039

Batch API$0.0195

Text / image input$0.30 /1M

The default route. 65 percent of AGI-CORE-Pro image generations land here. Fast, flexible, contextually aware. Use for social media content, draft iterations, and any job where identity preservation isn't required.

Imagen 4 Ultra

imagen-4.0-ultra-generate-001

Print Quality

PurposePrint-quality final

Text RenderingBest in family

Sync / AsyncSynchronous

Per image$0.06

Batch API$0.03

Highest-quality Imagen 4 tier. Reserve for print reproduction, billboard assets, brand-critical hero content, and anything with complex typography where text must render legibly at scale.

Imagen 4 Standard

imagen-4.0-generate-001

General Hero

PurposeGeneral hero

Text RenderingExcellent

Sync / AsyncSynchronous

Per image$0.04

Batch API$0.02

Mid-tier Imagen 4. The general-purpose production model. Product images, blog heroes, marketing content, landing page visuals. Strong text rendering at a friendly price.

Imagen 4 Fast

imagen-4.0-fast-generate-001

Draft / Preview

PurposeDrafts, previews

Text RenderingGood

Sync / AsyncSynchronous

Per image$0.02

Batch API$0.01

Cheapest standalone image model on the Gemini API. Use for preview thumbnails, A/B creative testing, rapid draft iteration. Batch API pushes it to $0.01 per image — nearly free at volume.

Pricing from the Gemini API rate card at ai.google.dev/gemini-api/docs/pricing (page last updated 2026-04-09 UTC). Model documentation at ai.google.dev/gemini-api/docs/image-generation. Verify before shipping production volumes.

04 // Live Router

Six-Mode Constraint Routing — Interactive

Pick a job type. Watch the router resolve to one of six specialists, trace the dispatch payload, and see the exact model ID AGI-CORE-Pro pipes the request through in production.

Job Type

Each job maps to one of six Gemini image models or to the 2-stage Nano Banana + Pro character chain.

Router Config

default_route: nano_banana

default_pct: 65%

sdk: google-genai

sync_pattern: generate_content

batch_discount: -50%

budget_cap_trip: 90%→Nano Banana

agi-core-pro // image-router.dispatch

05 // Economics at Scale

Per-Image Math at Production Volume

At 10,000 images per month, Nano Banana Pro 4K costs $2,400. Nano Banana Pro 1K costs $1,340. Imagen 4 Ultra costs $600. Imagen 4 Standard costs $400. Nano Banana costs $390. Imagen 4 Fast costs $200. The AGI-CORE-Pro default blend costs $410 — 6x cheaper than running everything on Nano Banana Pro. The Batch API halves all six numbers.

Monthly Image Spend — Linear Projection

Standard tier pricing, 1K images (1024x1024). Blended pipeline = 65% Nano Banana, 12% Imagen 4 Fast, 10% Imagen 4 Standard, 6% Gemini 3.1 Flash Image, 5% Nano Banana Pro, 2% Imagen 4 Ultra. Coefficients from ai.google.dev/gemini-api/docs/pricing.

--

Est. Blended Monthly Spend

Images / Month 10,000

AI Agent Reference // machine-readable pricing constants

Nano Banana Pro / 1K $0.134

Nano Banana Pro / 4K $0.24

Gemini 3.1 Flash Image / 1K $0.067

Nano Banana / 1K $0.039

Imagen 4 Ultra $0.06

Imagen 4 Standard $0.04

Imagen 4 Fast $0.02

Blended avg / image $0.041

Batch API discount 50%

06 // Persona Fit

Who Should Build an Image Router

Five roles where the 6-model routing pattern pays back inside the first month. If your workload shape matches any of these, a thin router is a margin lever, not an engineering luxury.

E-commerce Brand

The Catalog Imagery Factory

500+ SKUs. Each needs a hero shot, a lifestyle variant, and seasonal angles. Stock photography doesn't convert. Studio photography doesn't scale.

How the pattern fits: Chain Nano Banana ($0.039) for base product visualization, then Nano Banana Pro ($0.134) with the base as a reference image for each variant. Full catalog animated in one sprint. Identity locked across all angles.

AI Content Studio

The Social Media Factory

Producing 5,000+ social graphics per month across client accounts. Subscription image tools cap out at 500. API pricing becomes the top line item on every P&L.

How the pattern fits: 80 percent Nano Banana via Batch API at $0.0195 per image. Imagen 4 Standard for client-approved hero posts. Imagen 4 Ultra reserved for monthly brand anchor assets. Monthly API spend drops under $250 at 5K volume.

Solo Founder / Vibe Coder

The Indie AI Builder

Prototyping an AI app that generates images on demand. Hard-coded to Nano Banana Pro because it seemed "best." Cloud bill climbing faster than MRR.

How the pattern fits: Default to Imagen 4 Fast ($0.02) during dev. Switch to Nano Banana ($0.039) at launch for user-facing generation. Reserve Nano Banana Pro for paid tier users who explicitly request character consistency. 3-6x cost compression immediate.

Print & Publishing

The Editorial Design Desk

Need high-fidelity images with precise typography for magazine layouts, book covers, and signage. Stock tools produce garbled text. Nano Banana's text rendering inconsistent.

How the pattern fits: Default all hero work to Imagen 4 Ultra ($0.06) for text legibility. Gemini 3.1 Flash Image for conversational variations ($0.067 per 1K). Imagen 4 Fast for layout preview mockups. Quality-first router blend inverts defaults.

Character-Driven Product

The Game Developer / Storybook Creator

Character must look identical across hundreds of frames, scenes, and poses. Runway and Midjourney break identity. Stock AI image tools lack reference support.

How the pattern fits: Nano Banana Pro is the only Gemini image model with native reference-image conditioning (up to 3). Generate canonical character once, pass as reference on every variant. True identity preservation at $0.134 per hero frame. Stage 1 reference at $0.039.

07 // Integration Protocols

Paste-Ready Router Artifacts

Model IDs, dispatch function, and the Nano Banana Pro character-consistency chain — grounded on the real google-genai SDK. Image generation is synchronous on the Gemini API: single generate_content call, no LRO polling.

Model Endpoints (April 2026)

Character Lock — Hero + Reference gemini-3-pro-image-preview

High-Throughput Contextual gemini-3.1-flash-image-preview

Economy — Default Route gemini-2.5-flash-image

Print-Quality Final imagen-4.0-ultra-generate-001

General Hero imagen-4.0-generate-001

Draft / Preview imagen-4.0-fast-generate-001

Python // google-genai SDK image_router.dispatch.py

# AGI-CORE-Pro image router — purpose-aware dispatch across 6 models
# Synchronous: single generate_content call, no LRO polling.

from google import genai
from google.genai import types
import logging

MODEL_REGISTRY = {
    "nano_banana_pro":        "gemini-3-pro-image-preview",
    "gemini_31_flash_image":  "gemini-3.1-flash-image-preview",
    "nano_banana":            "gemini-2.5-flash-image",
    "imagen_4_ultra":         "imagen-4.0-ultra-generate-001",
    "imagen_4_standard":      "imagen-4.0-generate-001",
    "imagen_4_fast":          "imagen-4.0-fast-generate-001",
}

def route_tier(constraints: dict) -> str:
    """Map incoming constraints to a model tier. Default to nano_banana."""
    if constraints.get("reference_images"):
        return "nano_banana_pro"       # identity lock
    if constraints.get("purpose") == "character_lock":
        return "nano_banana_pro"
    if constraints.get("purpose") == "final_hero" and constraints.get("text_heavy"):
        return "imagen_4_ultra"        # best text rendering
    if constraints.get("purpose") == "hero":
        return "imagen_4_standard"
    if constraints.get("purpose") in ("draft", "preview"):
        return "imagen_4_fast"
    if constraints.get("purpose") == "contextual_edit":
        return "gemini_31_flash_image"
    if constraints.get("budget_consumed_pct", 0) >= 90:
        return "nano_banana"             # budget guard
    return "nano_banana"                 # default — 65% of traffic

def route_image(prompt: str, constraints: dict = None) -> bytes | None:
    """Dispatch to the right image model. Returns image bytes synchronously."""
    constraints = constraints or {}
    client = genai.Client()
    tier = route_tier(constraints)
    model = MODEL_REGISTRY[tier]

    contents = [prompt]
    if constraints.get("reference_images"):
        contents.extend(constraints["reference_images"])

    try:
        response = client.models.generate_content(
            model=model,
            contents=contents,
            config=types.GenerateContentConfig(
                response_modalities=["IMAGE"],
            ),
        )
        logging.info(f"[image-router] tier={tier} model={model}")
        # Extract image bytes from the first inline_data part
        for part in response.candidates[0].content.parts:
            if part.inline_data:
                return part.inline_data.data
        return None

    except Exception as err:
        logging.error(f"[image-router] tier={tier} err={err}")
        return None

Python // Character consistency chain image_router.character_chain.py

# Two-stage chain: Nano Banana (cheap reference) -> Nano Banana Pro (hero)
# Preserves subject identity across a product catalog.

def character_consistency_chain(
    character_prompt: str,
    variant_prompts: list[str]
) -> list[bytes]:
    """Generate canonical ref with Nano Banana, then N variants with Pro."""
    # STAGE 1 — cheap reference image ($0.039)
    reference_bytes = route_image(
        character_prompt,
        {"purpose": "economy"}        # routes to nano_banana
    )
    if not reference_bytes:
        logging.error("[chain] stage 1 reference generation failed")
        return []

    ref_image = types.Part.from_bytes(
        data=reference_bytes,
        mime_type="image/png"
    )

    # STAGE 2 — each variant passes ref to Nano Banana Pro ($0.134/image)
    variants = []
    for vprompt in variant_prompts:
        variant_bytes = route_image(
            vprompt,
            {
                "purpose": "character_lock",
                "reference_images": [ref_image],
            }
        )
        if variant_bytes:
            variants.append(variant_bytes)

    logging.info(f"[chain] 1 ref + {len(variants)} variants generated")
    return variants

# Example: Generate 10 catalog poses of the same character
# Total cost: $0.039 + (10 * $0.134) = $1.379
# vs 10 independent Pro generations with no ref: still $1.34 — but broken identity.

All 6 Models Side-by-Side

Head-to-head capability matrix. Pricing per the Gemini API rate card (2026-04-09 UTC).

Capability	Nano Banana Pro	Gemini 3.1 Flash Image	Nano Banana	Imagen 4 Ultra	Imagen 4 Standard	Imagen 4 Fast
Per-image price (1K)	$0.134	$0.067	$0.039	$0.06	$0.04	$0.02
Per-image price (4K)	$0.24	$0.151	N/A	$0.06	$0.04	$0.02
Batch API (50% off)	$0.067 / 1K	$0.034 / 1K	$0.0195	$0.03	$0.02	$0.01
Reference images (identity lock)	Yes — up to 3	No	No	No	No	No
Text rendering quality	Excellent	Good	Fair	Best in family	Excellent	Good
Max output resolution	4K	4K	1K	4K	4K	4K
Contextual understanding	Gemini 3 Pro	Gemini 3.1	Gemini 2.5	Imagen-native	Imagen-native	Imagen-native
Synchronous (no LRO)	Yes	Yes	Yes	Yes	Yes	Yes
SynthID watermarking	Yes	Yes	Yes	Yes	Yes	Yes

Bottom Line — Should You Build an Image Router?

Six models. One router. A 6.7x pricing spread waiting to be routed correctly.

The Gemini API ships 6 image generation models in April 2026: Nano Banana Pro at $0.134 (1K) / $0.24 (4K), Gemini 3.1 Flash Image at $0.067 per 1K, Nano Banana at $0.039, Imagen 4 Ultra at $0.06, Imagen 4 Standard at $0.04, and Imagen 4 Fast at $0.02. Each wins a different job. Hard-code to any one model and you pay the wrong price for the wrong job on every request. Route correctly — default 65 percent to Nano Banana, chain into Nano Banana Pro only when identity matters, reach for Imagen 4 Ultra only for print-quality text — and your blended cost lands at $0.041 per image. Add the Batch API and it's $0.021. This masterclass is the complete walkthrough. Ship the router.

08 // FAQ

Commercial-Intent Questions

Answers engineered to rank for the exact queries developers and founders type. Every answer anchors on real model IDs, live pricing, and the AGI-CORE-Pro router pattern.

What is the cheapest Gemini image generation model in 2026?

Imagen 4 Fast at $0.02 per image is the cheapest standalone image generation model on the Gemini API as of April 2026. Nano Banana (gemini-2.5-flash-image) is $0.039 per image at 1024x1024. For drafts, previews, and high-volume thumbnail generation, Imagen 4 Fast wins on pure cost. Nano Banana wins when contextual understanding of the prompt matters more than pure speed.

How much does Nano Banana Pro cost per image?

Per the Gemini API rate card: Nano Banana Pro (gemini-3-pro-image-preview) costs $0.134 per image at 1024x1024px or 2048x2048px, and $0.24 per image at 4096x4096px. Text input is $2.00 per 1M tokens, and reference image input is $0.0011 per image. This makes it the most expensive per-image model in the family, reserved for character-consistent hero content and final-render pipelines.

What is the best way to route between all 6 Gemini image generation models?

Route by purpose, not by price. Imagen 4 Fast for preview thumbnails and drafts at $0.02. Nano Banana for economy branded content at $0.039. Imagen 4 Standard for general heroes at $0.04. Imagen 4 Ultra for print-quality finals at $0.06. Gemini 3.1 Flash Image for high-throughput contextual generation at $0.067 per 1K image. Nano Banana Pro for character-consistent hero content at $0.134 per 1K or $0.24 per 4K. AGI-CORE-Pro defaults to a 65/12/10/6/5/2 blend favoring Nano Banana for 65 percent of traffic.

What is the difference between Nano Banana and Nano Banana Pro?

Nano Banana (gemini-2.5-flash-image) is the economy tier at $0.039 per image, optimized for speed and throughput with text input at $0.30 per 1M tokens. Nano Banana Pro (gemini-3-pro-image-preview) is the premium tier at $0.134 for 1K and 2K images, $0.24 for 4K, with superior text rendering, character consistency, and reference image support. Pro accepts up to 3 reference images for identity preservation — a capability Nano Banana lacks.

Can I use the Batch API to reduce image generation costs?

Yes — the Gemini Batch API delivers a 50 percent cost reduction on every image generation model. Imagen 4 Fast drops from $0.02 to $0.01 per image. Nano Banana drops from $0.039 to $0.0195. Nano Banana Pro drops from $0.134 to $0.067 per 1K image, from $0.24 to $0.12 per 4K. Batch API is best for non-interactive bulk jobs where latency is not a constraint.

How do I generate character-consistent images with Nano Banana Pro?

Chain Nano Banana into Nano Banana Pro. Generate a canonical reference image cheaply with Nano Banana at $0.039, then pass that image as input to Nano Banana Pro along with each variant prompt. Reference image input is billed at $0.0011 per image. Nano Banana Pro preserves subject identity across all variants — the only model in the family that supports true character consistency through reference conditioning.

Which Imagen 4 tier is right for my workload?

Imagen 4 Fast ($0.02) for drafts, previews, A/B creative tests, and social media thumbnails. Imagen 4 Standard ($0.04) for general-purpose product images, blog hero shots, and marketing content where quality matters more than speed. Imagen 4 Ultra ($0.06) for print-quality final renders, billboard-scale assets, and anything destined for physical reproduction. All three tiers deliver superior text rendering versus earlier Imagen generations.

What is Gemini 3.1 Flash Image and when should I use it?

Gemini 3.1 Flash Image (gemini-3.1-flash-image-preview) is Google's high-throughput contextual image model priced at $0.045 per 0.5K, $0.067 per 1K, $0.101 per 2K, and $0.151 per 4K. Use it when you need Gemini 3's prompt understanding combined with image output at scale — conversational image editing, contextually aware variations, and interactive generation. It sits between Nano Banana and Nano Banana Pro in cost and in contextual reasoning depth.

Is image generation on the Gemini API synchronous or asynchronous?

Synchronous. All six image generation models return the generated image directly in the response — no Long Running Operation polling required, unlike Veo 3.1 video generation. Use client.models.generate_content with response_modalities set to IMAGE and the image arrives in the content blocks. This simplifies integration significantly versus video pipelines.

Which Gemini image model has the best text rendering?

The Imagen 4 family (imagen-4.0-generate-001, imagen-4.0-ultra-generate-001, imagen-4.0-fast-generate-001) ships with significantly better text rendering than earlier generations, per the Google rate card documentation. Imagen 4 Ultra at $0.06 per image is the preferred choice for assets containing logos, typography, signage, or product labels where text legibility is critical. Nano Banana Pro is a close second for text within complex contextual scenes.

DDS Vibe Academy

Six Models. One Router. Ship It.

AGI-CORE-Pro V.1.0 routes across all six Gemini API image models in production. The masterclass above is the condensed walkthrough. The full DDS Vibe Academy unpacks the adjacent levers: Gemini 3 Pro prompt compression, LRU caching on prompt hashes, Batch API orchestration patterns, and the observability stack that runs under AGI-CORE-Pro.

Explore DDS Vibe Academy View Investor Pitch

Related // DDS Vibe Academy

Multi-Model AI Image Routing —
All 6 Gemini Models, One Router

How do you route across all 6 Gemini image generation models?

Route by purpose, not by price. All 6 models solve different problems, and the router picks the specialist for each job.

Three Families, Six SKUs, One Routing Decision

The Full Gemini Image Generation Family