Google shipped Nano Banana 2 Lite and Gemini Omni Flash on June 30, both available immediately through the Gemini API and Google AI Studio. The announcement was quiet by Google I/O standards — no keynote, no demo reel. The specs weren't quiet.

Nano Banana 2 Lite generates images from a text prompt in about four seconds. $0.034 per thousand images. Gemini Omni Flash generates and edits video conversationally — describe what you want, get a clip back, describe a change, get another clip — at $0.10 per second of output. A ten-second clip costs about a dollar. The two models are designed to work together: generate a still with Nano Banana 2 Lite, pass it to Omni Flash, animate it.

The ten-second ceiling on Omni Flash is real. Google isn't hiding it. Longer-form video requires stitching segments externally.

Source spread

• Google blog — Start building with Nano Banana 2 Lite and Gemini Omni Flash [hype] — official launch post. Claims best price-performance for video generation, frames conversational editing as the primary differentiator.
• Google Cloud Blog — Availability and pricing [builder] — API pricing, availability via AI Studio and Gemini Enterprise Agent Platform.
• WaveSpeed — Independent pricing breakdown [builder] — confirms $0.034/K images and $0.10/second; compares to Veo 3.1 family.
• VentureBeat — Enterprise video production angle [hype] — frames this as production-ready for enterprise workflows; leans on Google's framing throughout.
• Digen.ai — 2026 benchmark review [builder] — where the 40% character consistency improvement figure comes from; methodology is Google-sourced but independently reported.

Pros & cons

What works:

• The image pricing on Nano Banana 2 Lite is genuinely competitive. $0.034 per 1K images is cheaper than most Stable Diffusion API alternatives at comparable quality tiers. Four-second generation is production-viable for anything that isn't real-time.
• Conversational editing is the capability that actually differentiates Omni Flash from everything else in Google's own video lineup. Most video generation pipelines treat generation and editing as separate jobs requiring separate tools. Omni Flash's multi-turn architecture lets you describe what's wrong, get a corrected clip, describe what's still wrong — without starting over. That matches how creative direction actually works.
• A 40% improvement in character consistency versus 2025 AI video models is the metric that matters most for production use. Faces that drift shot to shot make video unusable. This is the right thing to measure.
• Available now, no waitlist, through the Gemini API. That's not a given in AI video generation right now.

What deserves scrutiny:

• Ten-second maximum output is a hard cap, not a soft limit. Longer-form video requires you to stitch segments. The seam quality at stitch points is not the same as native continuous generation — and no one is claiming it is.
• 720p output only. For consumer social content and thumbnails, fine. For broadcast or large-format display, it's a blocker until Google ships a higher-resolution option.
• Google's own Veo 3.1 Fast runs at the same $0.10/second at the same 720p resolution. What Omni Flash adds is the conversational editing and the multimodal input (text, images, and existing video as inputs). If you don't need those, Veo 3.1 Lite is cheaper.
• The 40% character consistency improvement is measured against "2025 AI models" without naming the specific comparison. First-party improvement claims should be treated as a starting point for your own evaluation.

What builders need to know

• The image-to-video pipeline is two API calls. Generate a still with Nano Banana 2 Lite ($0.034/K), pass the image to Omni Flash ($0.10/sec). A five-second clip with a generated still costs roughly $0.50. Factor this into your feature pricing before you promise it.
• 720p max is a real product constraint. If you need 1080p+, you're on Veo 3.1 Standard at $0.40/second — four times the cost. Know your resolution requirement before you route to Omni Flash.
• Test multi-turn session quality, not just first-clip quality. Google's claim is conversational editing. Your evaluation should include five-to-six-turn sessions checking whether the model maintains scene and character consistency throughout, not just on the first edit.
• Nano Banana 2 Lite is worth a direct eval against your current image pipeline. $0.034/K at four-second generation beats most alternatives on price. If you're paying more for Stable Diffusion API or a similar service, run the comparison.
• Both models wire into existing Gemini-based agent workflows through the Gemini Enterprise Agent Platform — no separate API key required if you're already using Gemini.

Further reading

• Google blog — Start building with Nano Banana 2 Lite and Gemini Omni Flash — official launch
• Google Cloud Blog — Availability and API pricing
• WaveSpeed — Independent pricing breakdown
• VentureBeat — Enterprise video production angle
• Gemini Omni Flash — model specs