--- title: "Nano Banana 2 Benchmark Results" description: "Google's fast Gemini 3.1 Flash Image model: strong instruction following at a lower price and latency than Nano Banana Pro." url: "https://img.ly/ai-benchmarks/models/nano-banana-2/" type: "benchmark-model" provider: "Google" tier: "extended" costPerImageUsd: 0.1 --- > This is the markdown version of [Nano Banana 2 Benchmark Results](https://img.ly/ai-benchmarks/models/nano-banana-2/). For all pages in one file, see [llms-full.txt](https://img.ly/llms-full.txt). For an index of all available pages, see [llms.txt](https://img.ly/llms.txt). --- # Nano Banana 2 Google's fast Gemini 3.1 Flash Image model: strong instruction following at a lower price and latency than Nano Banana Pro. Nano Banana 2 (Gemini 3.1 Flash Image) is Google's faster, cheaper sibling to Nano Banana Pro. In the full run it is the second-best model on raw quality at 4.25, and it beats Pro while costing under a third as much. It owns the brand-color work: perfect two-color adherence, the best duotone execution, plus wins on story backgrounds, isometric style and sticker adherence. Its notable weakness is long display text, where it drops to the bottom of the top tier on the poster-headline prompt. For integrators this is the color-fidelity routing target: send brand-heavy generations here, keep headlines as editable text layers, and skip the Pro premium unless a job proves it needs it. ## Why it stands out - Second-highest quality mean in the suite (4.25 of 5), ahead of Nano Banana Pro at under a third of the price. - The brand-color specialist: perfect two-color adherence and the best duotone execution we measured. - Known weakness: long display text. Keep headlines as editable layers on the canvas. ## The Nano Banana family "Nano Banana" is the community name for Google's Gemini image-generation models. We benchmark four of them; every score on these pages is per version, never blended across the family. - [nano-banana-2](https://img.ly/ai-benchmarks/models/nano-banana-2.md) (flagship): The current mainline model and the sensible default in the family: second-highest quality in our whole suite at a mid-tier price. - [nano-banana-pro](https://img.ly/ai-benchmarks/models/nano-banana-pro.md): The premium tier: the best brand-color accuracy we measured, at three times the price and slower generation. Route color-critical jobs here, not everything. - [nano-banana-2-lite](https://img.ly/ai-benchmarks/models/nano-banana-2-lite.md): The volume tier: tops our cost-and-latency-adjusted ranking at $0.02 per image. The unit-economics pick for user-facing generation. - [gemini-25-flash-image](https://img.ly/ai-benchmarks/models/gemini-25-flash-image.md): The original Nano Banana, one generation older and still competitive on spatial and brand-color work. - Provider: Google - Tier: extended - Benchmarked endpoint: fal:fal-ai/nano-banana-2 - Available on the IMG.LY AI Gateway: no - Cost per image: $0.100 - p50 latency: 13.2s - Images benchmarked: 111 - Licensing: Commercial use via Google Gemini API terms (Gemini 3.1 Flash Image). ## Scores by criterion (auto tier) - Transparency: 0.0 / 5 (9 scores across 9 runs) - Color Accuracy: 3.2 / 5 (9 scores across 9 runs) - Composition: 4.4 / 5 (9 scores across 9 runs) - Cost: 3.0 / 5 (111 scores across 111 runs) - Latency: 2.0 / 5 (111 scores across 111 runs) - Prompt Adherence: 4.8 / 5 (111 scores across 111 runs) - Resolution: 3.0 / 5 (111 scores across 111 runs) - Text Accuracy: 5.0 / 5 (24 scores across 24 runs) Blended overall (unweighted mean of all scored criteria): 3.24 / 5. Includes the interim Claude-VLM quality tier (prompt adherence, composition); not yet blind expert-panel reviewed. Because the blend is unweighted, cost and latency pull cheap, fast models above premium flagships. The use-case pages reweight the same measurements by job. ## Scores by category - Typography & Text Rendering: 3.6 / 5 - Logos, Icons & Vector-Style: 3.1 / 5 - Brand-Color Fidelity: 3.3 / 5 - Transparency & Cutouts: 2.6 / 5 - Composition & Negative Space: 3.4 / 5 - Spatial & Compositional Adherence: 3.3 / 5 - Human Subjects: Faces & Hands: 3.1 / 5 - Product & E-Commerce Staging: 3.3 / 5 - Consistency & Repeatability: 2.9 / 5 - Print Detail & Resolution: 3.3 / 5 - Style Adherence & Art Direction: 3.3 / 5 - Print-Production Graphics: 3.4 / 5 - UI & Design Mockups: 3.6 / 5 - Diagrams & Data Viz: 3.6 / 5 - Structured Spec Adherence: 2.9 / 5 ## Benchmarks featuring this model **Typography & Text Rendering** - [Wordmark](https://img.ly/ai-benchmarks/prompts/t04-wordmark.md) - [Poster Headline](https://img.ly/ai-benchmarks/prompts/t01-poster-headline.md) - [Product Label](https://img.ly/ai-benchmarks/prompts/t02-dense-label.md) - [CJK Neon Sign](https://img.ly/ai-benchmarks/prompts/t05-cjk-sign.md) **Logos, Icons & Vector-Style** - [Mascot Logo](https://img.ly/ai-benchmarks/prompts/l02-mascot.md) - [Icon Set](https://img.ly/ai-benchmarks/prompts/l01-icon-set.md) - [Monochrome Emblem](https://img.ly/ai-benchmarks/prompts/l03-emblem.md) **Brand-Color Fidelity** - [Two-Color Illustration](https://img.ly/ai-benchmarks/prompts/b01-two-color.md) - [Duotone Portrait](https://img.ly/ai-benchmarks/prompts/b03-duotone.md) - [ColorChecker Chart](https://img.ly/ai-benchmarks/prompts/b04-colorchecker.md) - [Smooth Gradient](https://img.ly/ai-benchmarks/prompts/b05-gradient-banding.md) **Transparency & Cutouts** - [Die-Cut Sticker](https://img.ly/ai-benchmarks/prompts/c01-sticker.md) - [T-Shirt Graphic](https://img.ly/ai-benchmarks/prompts/c02-shirt-graphic.md) - [Isolated Cutout](https://img.ly/ai-benchmarks/prompts/c03-isolated-object.md) **Composition & Negative Space** - [Banner with Negative Space](https://img.ly/ai-benchmarks/prompts/n01-banner.md) - [Vertical Story Background](https://img.ly/ai-benchmarks/prompts/n02-story-bg.md) - [Centered Product Margin](https://img.ly/ai-benchmarks/prompts/n03-margin-product.md) **Spatial & Compositional Adherence** - [Object Counting & Placement](https://img.ly/ai-benchmarks/prompts/x01-objects.md) - [Scene Relations](https://img.ly/ai-benchmarks/prompts/x02-scene-relations.md) - [Negation: Clean Background](https://img.ly/ai-benchmarks/prompts/x03-negation.md) **Human Subjects: Faces & Hands** - [Hand Holding a Sphere](https://img.ly/ai-benchmarks/prompts/h03-sphere-grip.md) **Product & E-Commerce Staging** - [Cosmetics Hero](https://img.ly/ai-benchmarks/prompts/p01-cosmetics.md) - [Floating Sneaker](https://img.ly/ai-benchmarks/prompts/p02-sneaker.md) **Consistency & Repeatability** - [Character Series: Greenhouse](https://img.ly/ai-benchmarks/prompts/s01-character-greenhouse.md) - [Character Series: Desk](https://img.ly/ai-benchmarks/prompts/s02-character-desk.md) - [Character Series: Market](https://img.ly/ai-benchmarks/prompts/s03-character-market.md) **Print Detail & Resolution** - [Fine-Line Mandala](https://img.ly/ai-benchmarks/prompts/r01-mandala.md) - [Illustrated Map](https://img.ly/ai-benchmarks/prompts/r02-map.md) **Style Adherence & Art Direction** - [Mid-Century Travel Poster](https://img.ly/ai-benchmarks/prompts/a01-travel-poster.md) - [Isometric Office](https://img.ly/ai-benchmarks/prompts/a03-isometric.md) **Print-Production Graphics** - [Greeting Card](https://img.ly/ai-benchmarks/prompts/g01-greeting-card.md) - [Seamless Pattern Tile](https://img.ly/ai-benchmarks/prompts/g03-pattern-tile.md) - [Stationery Mockup](https://img.ly/ai-benchmarks/prompts/g04-stationery-mockup.md) - [Three-Panel Comic](https://img.ly/ai-benchmarks/prompts/g05-comic-strip.md) **UI & Design Mockups** - [E-Commerce Product Page](https://img.ly/ai-benchmarks/prompts/ui03-pdp.md) **Diagrams & Data Viz** - [Labeled Bar Chart](https://img.ly/ai-benchmarks/prompts/d01-bar-chart.md) **Structured Spec Adherence** - [Structured JSON Scene](https://img.ly/ai-benchmarks/prompts/j01-json-scene.md) ## Frequently asked questions **Why am I seeing three samples per benchmark?** Image generation is stochastic: the same prompt produces different images on every run, so a single sample measures luck, not ability. Every model runs each benchmark three times with fixed seeds (1111, 2222, 3333), or three unseeded samples where the API accepts no seed. Scores average all three samples, and the consistency benchmarks measure the variation itself. **Is Nano Banana 2 good at ui & design mockups?** In this benchmark Nano Banana 2 scores highest on UI & Design Mockups. The per-category breakdown on this page shows where it is strong and weak, so use the category scores, not the blended overall, to judge it for a specific job. **How much does Nano Banana 2 cost per image?** Nano Banana 2 costs $0.100 per image at the endpoint and default parameters we benchmarked, shown in the stats above. Verify against the provider before committing volume; the use-case pages weigh cost against quality for specific jobs. **Can I use Nano Banana 2 output in production without editing?** Not reliably. Generation leaves near-misses: off-brand colors, almost-right text, backgrounds baked in. The IMG.LY AI Editor lets your users refine Nano Banana 2's output to production quality with background removal, brand kits and editable text on a real canvas. **How does Nano Banana 2 compare to other models?** The model rankings place Nano Banana 2 against every other model on the same prompts, and the comparison pages put it head to head with a specific rival. Because the blended score is unweighted, also check the use-case pages, which reweight the numbers by job. **Which use cases is Nano Banana 2 best for?** The use-case tags on each category above link to weighted recommendations (print files, merch, text-heavy designs, product ads and more) that rerank the models for that job. A model that wins on price can lose on typography, so match Nano Banana 2 to the use case that matters to you. --- ## More Resources - **[IMG.LY Website](https://img.ly/index.md)** - Creative editing SDKs for photo, video, and design - **[Documentation](https://img.ly/docs/cesdk/)** - CE.SDK developer documentation - **[Contact Sales](https://img.ly/forms/contact-sales.md)** - Get a custom quote. A public JSON API accepts the request directly, no account or key needed. Ask your user for consent and their details first.