HomeAI models
Google: Gemini 3.1 Flash Image
Google
•
0
•
Released 
February 2026

Google
0

Pro-level image generation and editing at Flash speed, with web-grounded rendering and multilingual in-image text.

Modality:
Text
Image
Video
PDF
model ID
google/gemini-3.1-flash-image
Start for free
arrow black icon
View API Docs
arrow white icon

Output Speed *

0.00
tok/s

Intelligence Index *

0
/ 100

Context Window *

0
tokens

Input price

3
Anytoken

Output price

18
Anytoken

Pro-Grade Image Generation at Flash Speed

‍
Gemini 3.1 Flash Image, publicly known as Nano Banana 2, is Google DeepMind's Flash-tier image generation and editing model. It sits below Nano Banana Pro (Gemini 3 Pro Image) in the lineup but combines much of that quality with faster, cheaper inference. Built on Gemini 3 Flash, it accepts text, image, and PDF input and returns generated or edited images plus text. It pulls from Gemini's world knowledge and web search to render accurate subjects, legible multilingual in-image text, and consistent characters. The best fit is high-volume generation and iterative editing pipelines that need production speed without collapsing quality.

Start building with Gemini 3.1 Flash Image via the AnyAPI.ai API.

Performance

Where Nano Banana 2 Earns Its Place in a Production Pipeline

Gemini 3.1 Flash Image is built for fast, repeatable generation and editing rather than maximum fidelity. It maintains visual coherence across up to five characters and fourteen objects in a single workflow, renders legible multilingual in-image text, and grounds subjects using Google Search. Independent comparisons report roughly 2–3x faster generation than Nano Banana Pro at materially lower cost, while retaining most of Pro's quality. For teams generating large image volumes or iterating quickly, that speed-to-quality ratio removes the hesitation that slower, pricier models introduce, making it a strong everyday default.

Benchmarks

How Nano Banana 2 Compares in Independent Testing

Google has not published formal quantitative image benchmarks for this model, so most evidence is directional. Independent comparisons consistently place Nano Banana 2 at roughly 2–3x the generation speed of Nano Banana Pro, with several reviewers estimating around 90–95% of Pro's image quality in typical scenarios. Third-party sources also report it ranking at the top of public text-to-image arenas at launch. Treat these figures as editorial estimates from reviewers rather than official measurements; they align with Google's own Flash-speed positioning but were not produced under controlled, published benchmark conditions.

Output Speed

*
0.00
tok/s

Intelligence Index

*
0
/ 100

MMLU *

Broad world knowledge and problem-solving
0
%

GPQA *

PhD-level scientific reasoning across physics, biology, chemistry.
0
%

HLE *

Adherence to multi-step structured instructions.
0
%

LiveCodeBench *

Tool-calling reliability in long agentic loops.
0
%

Technical Specifications

What the model supports

Gemini 3.1 Flash Image accepts text, image, and PDF input and returns images plus text. It supports a 131,072-token input context and outputs images up to 4K across 14 aspect ratios, alongside up to a 64K text output. Thinking is configurable, and Google Search grounding (including image grounding) is available. The two specifications with the greatest production impact are the resolution flexibility (0.5K–4K) and web grounding, which together let one endpoint serve draft-quality previews and near-final assets while rendering accurate real-world subjects.
Verified Specifications — 
0
*
Input modalities
Text
Image
Video
PDF
output modalities
Text
Image
Context window
0
 tokens
Maximum output tokens
0
Reasoning
No
Knowledge cutoff
February 2026
Pricing (standard)
3
 AnyTokens in
 / 
18
 AnyTokens out

Limitations & Trade-offs

Where 0 falls short

1
Fidelity ceiling below Pro. Gemini 3.1 Flash Image trades some quality for speed. Independent reviewers estimate it reaches roughly 90–95% of Nano Banana Pro's quality, with Pro holding a modest edge in 4K texture detail and natural lighting. This matters for high-stakes final assets like packaging, large-format print, or portfolio work, where Nano Banana Pro remains the safer choice. For drafts and high-volume web content, the gap is rarely decisive.
2
Text rendering limitations. Google's own model card notes text rendering is poor for small text (often blurry at 1K), long paragraphs, and page-length copy. Character consistency between input and generated images is not always perfect, and masked/doodle-based editing only partially follows instructions. Applications that depend on precise typography, dense labels, or UI mockups with fine print should validate output carefully or prefer Nano Banana Pro, which offers stronger text accuracy.
3
Grounding-dependent factuality. The model's knowledge cutoff is January 2025, and Google's card lists world knowledge, 3D reasoning, and factuality as still-limited areas. Accurate renderings of recent or specific real-world subjects rely on Google Search grounding rather than the base model. Workloads requiring up-to-date or factually precise visuals must enable grounding and still verify results; do not treat generated imagery as authoritative reference.
4
Mandatory SynthID watermarking. Every image created or edited with Gemini 3.1 Flash Image carries an invisible SynthID watermark identifying it as AI-generated. This supports provenance and transparency but cannot be disabled. Teams needing unmarked output for specific licensing or downstream requirements should account for this, as it is a fixed characteristic of all Nano Banana 2 outputs.

Best-Fit Workloads

Where this model earns its place

01

High-volume image generation pipelines

‍
The model is explicitly positioned for creating visuals at scale with strong price-performance. Its faster inference and lower cost per image versus Nano Banana Pro make it economical for generating thousands of assets for social media, marketing variations, or catalog imagery. Multiple resolution tiers (0.5K–4K) let you match output size to purpose, keeping cost down for previews while reserving higher resolution for final assets.

02

Iterative image editing and refinement
‍

Nano Banana 2 is built for rapid edits and iteration: adjusting elements, restyling, and regenerating variations quickly. Its speed invites experimentation that slower models discourage, making it well suited to design exploration and conversational editing loops. Note that masked/doodle editing only partially follows instructions and character consistency between input and output is imperfect, so precise, high-stakes edits may still warrant Pro.

03

Infographics, diagrams, and web-grounded visuals
‍

Because it draws on Gemini's world knowledge and real-time web search, the model can render accurate subjects, turn notes into diagrams, and build infographics and data visualizations. This suits editorial, educational, and internal-tooling use cases where visuals must reflect real-world information. Enable Search grounding for accuracy, and verify factual details given the January 2025 base knowledge cutoff.

04

Multi-character storyboards and product scenes

‍
The model maintains visual coherence for up to five characters and fourteen objects in a single workflow, which supports storyboards, narrative sequences, and multi-product recontextualization. Combined with 14 aspect ratios, it fits app flows that assemble consistent scenes across frames. For final, brand-critical compositions where consistency must be exact, validate output or escalate to Nano Banana Pro.

Pricing in anytokens via AnyAPI
Input
3
₳
Output
18
₳
Cache write
—
₳
Cache read
—
₳
Get API access
arrow black icon
Read Docs

on this page

PerformanceBenchmarksTechnical SpecificationsLimitations & Trade-offsBest-Fit WorkloadsIntegrationFAQ

Integration

Access 0 via AnyAPI.ai

Access 0 through AnyAPI.ai using a unified API built for multi-model AI applications. Integrate 0 without maintaining a separate provider-specific connection, and keep the flexibility to test, switch, or combine models as your application requirements evolve.

01

One API integration

Access 0 and other AI models through the same API workflow instead of maintaining separate integrations for every provider.

02

Easy model switching

Test 0 against alternative models or switch models as your performance, capability, or cost requirements change without rebuilding your application around another provider API.

03

Flexible for production

Use 0 from experimentation through production while keeping your AI stack flexible as workloads, traffic, and model requirements evolve.

04

Multi-model applications

Use 0 for the workloads where it performs best and combine it with other models for tasks that require different capabilities, performance, or efficiency.

Start free
arrow black icon
Read Docs

Frequently Asked Questions

Answers to common questions about integrating and using this AI model via AnyAPI.ai

Gemini 3.1 Flash Image, known publicly as Nano Banana 2, is Google DeepMind's Flash-tier image generation and editing model, released in February 2026. Built on Gemini 3 Flash, it delivers near-Pro image quality at faster speed and lower cost. It handles text-to-image, image editing, and multi-image composition, and returns images plus text.

Gemini 3.1 Flash Image supports a 131,072-token input context window. This accommodates detailed prompts, reference images, and multi-turn generation sessions. On the output side it can produce images up to 4K resolution across 14 aspect ratios, plus text output up to 64K tokens.

Nano Banana 2 (Gemini 3.1 Flash Image) is speed- and cost-optimized, while Nano Banana Pro (Gemini 3 Pro Image) is quality-optimized. Independent reviewers report Nano Banana 2 generating roughly 2–3x faster at around 90–95% of Pro's quality. Choose Nano Banana 2 for high-volume iteration and Pro for maximum fidelity and text precision.

Yes. Gemini 3.1 Flash Image supports grounding with Google Search, including grounding for images, allowing it to render more accurate subjects using real-time information. Its base knowledge cutoff is January 2025, so grounding is the recommended way to reflect recent or specific real-world details in generated visuals.

Yes. Every image created or edited with Gemini 3.1 Flash Image includes an invisible SynthID digital watermark identifying it as AI-generated, supporting provenance and transparency. The watermark is applied to all outputs and is part of Google's standard safety and authenticity features for the model.

* Benchmark data source: Artificial Analysis artificialanalysis.ai