Google: Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)

Create a clean product photo of a ceramic coffee mug on a wooden table with soft morning light. Minimalist composition, warm tones, shallow depth of field.

The model generates a photorealistic image showing a white ceramic mug centered on light oak planks. Sunlight streams from the left, casting a soft shadow and creating a gentle gradient across the mug's surface. The background blurs into warm cream tones. The composition feels commercial-ready, with accurate material rendering—the ceramic has subtle surface variation and the wood grain shows natural texture. Color temperature leans warm without oversaturation.

This example highlights Nano Banana's strength in product photography with natural lighting. The 65k token context window allows detailed scene description in a single prompt. However, at $1.50/Mtok output, generating multiple variations for client review adds up quickly compared to dedicated product-shot models.

Illustrate a technical diagram showing how a heat pump works. Include labeled components, directional arrows for refrigerant flow, and a cutaway view. Engineering blueprint style.

The model produces a clean technical illustration with a cross-section of a heat pump system. Components—compressor, condenser, evaporator, expansion valve—are clearly delineated with thin black outlines. Blue and red arrows trace refrigerant flow through the cycle. Labels appear in a sans-serif font with leader lines. The style resembles patent drawings: precise, uncluttered, with just enough shading to convey depth. Background is off-white with a subtle grid.

Demonstrates the model's capability for structured, informational graphics where accuracy matters more than artistic flair. The Flash Lite architecture handles geometric precision well. Trade-off: complex multi-panel diagrams sometimes require iteration to get label placement right, and each regeneration costs at the output rate.

Design a whimsical book cover illustration for a children's story about a fox who opens a bakery. Bright colors, hand-drawn feel, friendly and inviting.

The image shows an orange fox wearing a white apron, standing behind a wooden counter piled with pastries. The fox has large, expressive eyes and a welcoming smile. Behind it, shelves display bread loaves and cakes in saturated yellows, pinks, and blues. The style mimics gouache painting—visible brushstrokes, slightly uneven edges, warm lighting. A chalkboard menu hangs on the wall. The overall mood is cozy and approachable, suitable for ages 4-8.

Showcases the model's range beyond photorealism—it handles illustrative styles with character consistency and appropriate tonal warmth. The Gemini 3.1 base handles stylistic instructions reliably. Limitation: fine text on the chalkboard often needs manual cleanup, and the $0.25 input cost means long style-reference prompts aren't free.

When Flash Lite handles high-volume receipt OCR under budget

A 12-person consulting firm processes 400-600 expense receipts monthly through Slack uploads. Gemini 3.1 Flash Lite Image hits the sweet spot here: $0.25/Mtok input means each receipt costs roughly $0.0003 to parse (assuming 1.2k tokens per image), totaling under $0.20/month at peak volume. The 65k context window lets you batch 15-20 receipts in a single call when reconciling weekly, cutting API overhead. Output is structured JSON (vendor, date, total, line items) piped straight into QuickBooks. If you're over 2,000 receipts/month or need sub-second latency, step up to the standard Flash model—but for typical team expense flows, this is the cheapest reliable option that doesn't choke on crumpled photos.

Flash Lite as first-pass filter before human review

An 8-person e-commerce team reviews 200 user-submitted product photos daily for policy violations (nudity, weapons, counterfeit logos). Flash Lite runs as the initial gate: each image gets a binary pass/fail and a two-sentence explanation in under 2 seconds. At $0.25 input + $1.50 output per Mtok, the daily cost is around $0.15 (6k tokens/day average). Flagged images (roughly 12%) route to human moderators; clean ones publish immediately. The 65k context window is overkill for single-image calls, but you can batch 10 images with shared instructions to halve request overhead. If your false-negative tolerance is under 2%, you need a heavier model with public safety benchmarks—but for high-throughput pre-filtering where humans catch edge cases, this is the right price-speed trade.

When Flash Lite gives fast UI critique without designer-level cost

A 5-person product team iterates on Figma wireframes 3-4 times per feature sprint. They screenshot each frame (mobile + desktop views) and ask Flash Lite for accessibility checks, layout consistency notes, and copy readability flags. Each critique costs $0.0008 (3k tokens in, 2k out), so a 12-screen feature review runs under $0.01. The model catches obvious issues—contrast ratios, button alignment, truncated labels—in 3-5 seconds per screen, letting designers focus review time on interaction flows and brand fit. The 65k context window means you can feed all 12 screens in one call with a shared style guide, though latency creeps past 10 seconds. If you need pixel-perfect design system enforcement or brand voice scoring, upgrade to a multimodal reasoning model—but for fast iteration feedback, Flash Lite keeps sprint velocity high without burning budget.