Nano Banana 2 Lite + Gemini Omni Flash Just Shipped: $0.034 / 1K Images and $0.10 / sec Video
Google released two new Gemini media models for developers — Nano Banana 2 Lite at $0.034 per 1,000 images with 4-second latency, and Gemini Omni Flash at $0.10 per second of video in public preview. Here's what shipped, the pricing math, and how to integrate.
Google shipped two new generative media models for developers today — both with sharper pricing than the previous tier and both targeting the high-throughput, multi-tenant workloads that make per-call economics actually matter. Nano Banana 2 Lite (the new low-end of Google’s image family, internally gemini-3.1-flash-lite-image) lands at $0.034 per 1,000 images with ~4-second latency. Gemini Omni Flash (the production-grade Omni variant we’d previously seen in leak form) hits public preview at $0.10 per second of video output with native synced audio.
Together they round out a complete Gemini-native creative pipeline: Lite for cheap, fast image generation, and Omni Flash to animate those images into video with conversational editing. Below is what actually shipped, the pricing breakdown, and how it fits against the rest of the frontier.
What shipped, in one table
| Detail | Nano Banana 2 Lite | Gemini Omni Flash |
|---|---|---|
| Model ID | gemini-3-1-flash-lite-image | gemini-omni-flash-preview |
| Status | GA | Public preview |
| Pricing | $0.034 / 1,000 images | $0.10 / sec of video |
| Latency | ~4 seconds per image | (varies by clip length) |
| Inputs | Text + image | Text + image + video |
| Outputs | Image | Video + synced audio |
| Modalities | Image generation, editing | Conversational video editing, T2V, I2V, reference-to-video |
| Distribution | AI Studio, Gemini API, Enterprise Agent Platform, AI Mode in Search, Gemini app | AI Studio, Gemini API, Enterprise Agent Platform |
| Watermarking | C2PA + SynthID (default, non-optional) | C2PA + SynthID (default, non-optional) |
| Provisioned throughput | Available today | Rolling out soon |
Three things to pull out of that table:
- $0.034 per 1,000 images is the cheapest serious image model on the market. That’s $0.000034/image. For reference, GPT Image 2’s lowest tier is roughly an order of magnitude higher; legacy Nano Banana (Gemini 2.5 Flash Image) was about 6× this cost.
- $0.10/sec for native audio-video output is competitive with mid-tier Seedance/Wan/Kling pricing but lower than Veo 3.1 — and it’s a public-preview number, meaning Google is anchoring for production launch rather than testing pricing-elasticity in the wild.
- C2PA + SynthID are non-optional. Same posture as the prior Omni Flash leak coverage: this is consumer-facing first, developer-facing second. There’s no API knob to disable watermarking. Commercial use cases that require unwatermarked output won’t find a path here.
Nano Banana 2 Lite: what changed vs the rest of the family
Google’s image family now stacks like this:
| Tier | Model ID | Use case | Approx pricing |
|---|---|---|---|
| Lite (new) | gemini-3-1-flash-lite-image | High-volume / cheap | $0.034 / 1K images |
| Standard | Nano Banana 2 (gemini-3-flash-image) | Production-grade | mid-tier |
| Pro | Nano Banana Pro (gemini-3-pro-image) | High-fidelity / 4K | premium |
The Lite tier specifically targets:
- A/B testing ad variations at scale — when you need 50–500 variants per campaign and unit cost dominates
- Social-media apps that generate per-user creative assets where you can’t pass on premium-tier costs
- Background/auxiliary imagery in larger pipelines (banners, listing thumbnails, placeholder art) that doesn’t need Pro-tier 4K
- Internal eval pipelines generating reference clips at volume
What it gives up vs Nano Banana 2 (Standard): some peak fidelity, slightly weaker text-in-image rendering, and 4K-output ceiling. Practical quality is reportedly close enough that Lite covers most use cases where you’d previously have shipped the Standard tier.
The headline framing from Google’s own blog: “a significant leap in visual quality versus the legacy Nano Banana (Gemini 2.5 Flash Image).” In plain terms — Lite is the new floor, and it’s higher than the old floor.
Gemini Omni Flash: now actually available
When Gemini Omni Flash shipped to the Gemini app last month, the developer API was “coming weeks.” That window is now closed. It’s live on AI Studio and the Gemini API today at $0.10/sec.
Public-preview status means three things in practice:
- Pricing can move before GA. Anchor your cost models cautiously.
- Rate limits are tighter than they’ll be at GA. High-volume workloads should plan around current ceilings.
- Provisioned throughput “rolling out soon” — until that lands, sustained high-concurrency calls aren’t supported.
The capability surface that shipped:
- Multi-modal inputs: text + image + video reasoned across in a single prompt
- Conversational editing: character swap, relighting, camera-angle changes through chat instructions
- Native synced audio generated alongside video
- World knowledge for physics, text rendering, action consistency
- Character + object consistency across edits
Coming-soon list (announced but not shipped):
- Audio references
- Video references
- Last-frame control
- Scene extension
- Higher resolutions
That coming-soon list maps closely to the production video-model feature set Seedance 2.0 and Wan 2.7 already ship — so the gap will close at GA, not at preview.
The combined-pipeline angle
The most interesting framing in Google’s announcement is the two-step creative workflow they’re suggesting:
prompt → Nano Banana 2 Lite (image at $0.000034) → Gemini Omni Flash (video at $0.10/sec)A 5-second video clip with one source image lands at: $0.000034 + (5 × $0.10) = $0.500034. That’s the all-in cost for a complete image-to-video creative — and the image generation is essentially free in this math.
For applications that need to ship per-user creative content (UGC apps, generative ads, personalized video greetings), that workflow’s unit economics finally work at a scale that didn’t make sense with Veo 3.1 + DALL-E pricing six months ago.
Where this fits in the broader landscape
Two adjacent points:
Against image-gen frontier: Nano Banana 2 Lite at $0.034/1K images undercuts the previous low-end (Mai Image 2.5 Standard, Flux 2 Schnell, Seedream 4 Mini) by 30-80% depending on which you measure against. For volume creative-asset workloads it’s the new default-to-evaluate option.
Against video-gen frontier: Gemini Omni Flash at $0.10/sec sits in the middle of the production-grade video pack. Cheaper than Veo 3.1 (~$0.50/sec), cheaper than Sora 2 high-tier, comparable to Seedance 2.0 (Standard), more expensive than the new tier of disruptors (Agnes Video V2.0 at $0.005/sec, Seedance 2.0 Mini reported well below Fast). The differentiator vs the cheap tier isn’t price — it’s the multi-modal input + conversational editing surface that requires Gemini’s LLM backbone.
How to access today
Three deployment paths depending on what you’re building.
Through Google directly
gemini-3-1-flash-lite-imageis GA on AI Studio, Gemini API, Vertex AI, and Gemini Enterprise Agent Platformgemini-omni-flash-previewis in public preview on AI Studio and Gemini API- Both models also surface on consumer Google products (AI Mode in Search, Gemini app)
Through WaveSpeedAI
WaveSpeedAI hosts the broader Google model family (Nano Banana 2, Nano Banana Pro, Veo 3.1, Veo 3 Fast) alongside the rest of the frontier under one API key, so you can A/B test new tiers against your existing model lineup without rotating provider credentials.
If you’re building consumer-facing image generation, the WaveSpeedAI image generator gives you Nano Banana 2 and the Pro variants directly through a hosted UI — useful when you want to validate quality on real prompts before wiring API integration.
If you’re building video, the WaveSpeedAI video generator covers Seedance, Veo, Sora, Kling, and Wan under the same interface. When Nano Banana 2 Lite and Gemini Omni Flash propagate into the platform, you’ll be able to compare them against the existing lineup under the same surface.
Through agent routers
Both new model IDs are also propagating through aggregators (Vercel AI SDK, LangChain, etc.). Routing policies like “use Nano Banana 2 Lite for high-volume image generation, fall back to Nano Banana Pro for 4K hero shots” become one-line config changes once the model IDs land in your router’s registry.
What to watch for in the next two weeks
Three signals:
- Provisioned throughput pricing for Omni Flash. When PT lands, that defines the unit economics for high-concurrency production use. Watch for a per-token-second commitment rate that converts the preview pricing to predictable monthly cost.
- Replication of Google’s “4-second latency” claim under real load. First-party latency numbers are at idle capacity. The interesting question is what p50/p95 look like at production concurrency — and how PT changes that curve.
- GA timeline for Omni Flash. Public preview → GA usually takes 4–12 weeks for Google’s recent media models. When GA pricing publishes, that resets your cost model for production planning.
Until then: the Nano Banana 2 Lite + Gemini Omni Flash combo is the cleanest price-to-capability story Google has shipped this year, and it’s the right time to set up your A/B against your current image+video pipeline.
Sources: Google Cloud blog announcement, Google blog (consumer), VentureBeat coverage, SiliconAngle, 9to5Google on NotebookLM integration.