Nano Banana 2 Lite & Gemini Omni Flash (Google)

Google's launch post for two generative-media models in the gemini line — the first time this spoke has concrete IDs, latency, and per-unit pricing for the family's image and video tiers (the May 2026 roundup only had marketing positioning, google-ai-updates-may-2026). T3 vendor blog; numbers are launch claims, treat as dated snapshots.

Nano Banana 2 Lite — text-to-image

• Model ID gemini-3.1-flash-lite-image; the speed/cost tier of Google's Nano Banana image line, and the replacement for the first-gen Nano Banana.
• ~4-second generations at $0.034 per 1,000 images, pitched for "rapid ideation and high-velocity developer pipelines where speed and cost are the primary constraints."
• Claims strong prompt adherence, character consistency, and legible in-image text — Google shows it on an Elo-vs-latency-vs-cost trade-off chart against competitors (the quality axis is an Elo score, same shape as the text llm-benchmarks).

Gemini Omni Flash — video generation & editing

• Model ID gemini-omni-flash-preview; the cost-efficient tier of the Gemini Omni line, for video generation and conversational (natural-language) editing, with multimodal referencing (text + image
  • video inputs).
• $0.10 per second of video output — explicitly "the same as Veo 3.1 Fast," so Google is pricing Omni Flash at parity with its existing fast video model. 10-second max generation length.
• Preview limits stated honestly: no audio references, no API scene-extension, video refs up to 3s accepted but not processed correctly, character consistency wobbles across scene changes.

The intended workflow

Google pitches the two together: generate a still with Nano Banana 2 Lite, then pass it as a reference to Omni Flash to animate it into video — an image→video pipeline, with the Interactions API supporting up to three sequential edits. Both ship on AI Studio, the Gemini API, and the Gemini Enterprise Agent Platform, rolling out to consumer surfaces (AI Mode in Search, the Gemini app, Google Flow).

Why it's here, and a boundary note

These sit in gemini as the family's generative-media tiers — the provider/model market this spoke owns. They also add a new pricing unit to the spoke ($/image, $/second-of-video) alongside the $/token text tiers in llm-api-pricing. Boundary watch: image/video generation is a distinct modality from the text/multimodal LLMs at this spoke's core (and audio-gen was already carved out to speech-audio-wiki). For now it routes here as part of the Gemini family; if non-Google generative-media sources accumulate, a dedicated image/video-gen spoke is the spin-out candidate (see synthesis).

Prices re-verified 2026-08-30, and kept volatile deliberately

Both launch figures still hold two months on: $0.034 per 1,000 images for Nano Banana 2 Lite and $0.10 per second of output for Omni Flash. Two additions the launch post did not carry:

• The image price resolves more precisely as $0.0336 per 1K-resolution image at standard rates, with $0.0168 batch — a straight 50% batch discount, the same lever the text tiers use (llm-api-pricing). Generative media is being sold on the same commercial machinery as tokens, not on a separate one.
• Nano Banana 2 Lite also carries token pricing — $0.25 / $1.50 per million in/out — so a single model is billed per-image and per-token depending on how it is called. That complicates the "new pricing unit" note below: the unit did not replace tokens, it was added beside them.

This page stays volatile on purpose, against the 2026-08-04 convention that a dated publication is a snapshot and should be stable. It is the clearest exception in the corpus: a launch post whose per-image and per-second figures are the current rate card, actionable today, and the reason anyone reads this page. Snapshot grading would freeze a live price. It will surface on the stale queue again in 60 days, which is the convention working rather than failing.

Related

gemini · google · google-ai-updates-may-2026 · llm-api-pricing · llm-benchmarks · synthesis