On June 30, Google shipped two APIs that together let you build a complete AI media pipeline — still image to animated video — for under a dollar per asset. Nano Banana 2 Lite generates images in 4 seconds at $0.034 per 1,000. Gemini Omni Flash turns those images into 10-second videos at $0.10 per second. Neither model is the most capable thing in Google’s lineup. Together, they’re the most practical thing Google has shipped for developers building media products in 2026.
Nano Banana 2 Lite: The Lite Label Undersells It #
The “Lite” in Nano Banana 2 Lite suggests a compromise model. The benchmark numbers say otherwise. On the Text-to-Image Arena, Nano Banana 2 Lite scores an Elo of 1,251 — beating Nano Banana Pro’s 1,245. The faster, cheaper model is outperforming the one above it on the pricing sheet.
The specs are worth laying out clearly. At 1K resolution, each image costs $0.000034. Generation takes roughly 4 seconds. The model ID is gemini-3.1-flash-lite-image
, accessible through the Gemini API, AI Studio, Vertex AI, and Firebase. Logan Kilpatrick, Google AI Studio lead, framed the speed directly: “The speed of Nano Banana 2 Lite is going to enable so many new use cases where there is a high degree of latency sensitivity.” At 4 seconds, you can put image generation directly in an interactive user flow — not just a batch job.
The ceiling matters too. Nano Banana 2 Lite tops out at 1K resolution. If you need 4K for print or high-fidelity production, step up to the full Nano Banana 2 at gemini-3.1-flash-image
($0.067 per 1K image, same API pattern). For web, ads, e-commerce thumbnails, and social graphics, the Lite model handles it at half the cost.
Gemini Omni Flash: Video Generation as a Conversation #
Gemini Omni Flash (gemini-omni-flash-preview
) is the more architecturally interesting API. This is not a submit-and-wait video generator. It uses the Interactions API, which keeps session state across requests. You generate a video, examine it, send a follow-up prompt — “change the background to gradient blue” — and get a revised version without starting over. Up to three sequential edits per session, threaded via previous_interaction_id
.
Pricing lands at $0.10 per second of video output, level with Veo 3.1 Fast and Sora 2’s 720p tier. Current output caps at 10 seconds and 720p. Audio references and scene extension aren’t in the API yet. Character consistency can degrade during scene changes. This is a public preview, and the behavior reflects that.
What Gemini Omni Flash does well right now: background swaps, style transfers, simple object replacements, and product animations. Tao Zhang at Manus AI is already running it inside autonomous agent workflows where the conversational edit loop is the point, not a workaround.
The Pipeline: Image to Video, $0.53 Per Asset #
Here’s where the two models make a case together. A typical product creative workflow:
- Generate a product still with Nano Banana 2 Lite: ~4 seconds, ~$0.034
- Animate it with Gemini Omni Flash (5-second clip): ~$0.50
- Refine with one conversational edit if needed: ~$0.50
Total for a complete image + video creative: roughly $0.53 on the first pass. At 1,000 SKUs, that’s around $530 for a full creative suite — still images and short-form video for every product. One API key, one SDK, one billing relationship. Google embeds SynthID watermarks in both image and video outputs, so provenance tracking is consistent across the pipeline.
Where This Fits Against the Competition #
Teams evaluating Sora 2 should note the API sunsets September 24, 2026. Gemini Omni Flash at the same $0.10/second price point is the clearest migration path if conversational editing matters to your workflow. For pure cinematic quality, Runway Gen-4.5 still leads on control. Veo 3.1 Standard offers 1080p at $0.40/second for teams where resolution matters more than edit speed.
| Model | Price | Max Resolution | Best For |
|---|---|---|---|
| Nano Banana 2 Lite | $0.034/1K img | 1K | Web, ads, real-time |
| Nano Banana 2 (std) | $0.067/1K img | 4K | Production, print |
| Gemini Omni Flash | $0.10/sec | 720p | Quick video, iteration |
| Veo 3.1 Standard | $0.40/sec | 1080p | High-quality video |
Getting Started #
Both models are live. Start with the Gemini API image generation docs for Nano Banana 2 Lite and the Omni Flash docs for video. Google AI Studio is the fastest way to test both without infrastructure setup. The Python pattern for the combined pipeline:
from google import genai
client = genai.Client()
image = client.interactions.create(
model="gemini-3.1-flash-lite-image",
input="Product shot, wireless headphones, white background, studio lighting",
config={
"response_format": {
"type": "image",
"image_size": "1K",
"aspect_ratio": "1:1"
}
}
)
video = client.interactions.create(
model="gemini-omni-flash-preview",
input="Product rotating 360 degrees, white background, studio lighting"
)
refined = client.interactions.create(
model="gemini-omni-flash-preview",
input="Add a soft gradient blue background",
config={"previous_interaction_id": video.id}
)
Preview status means pricing and behavior are subject to change — plan fallback paths for anything involving faces, brands, or accuracy-sensitive claims. Neither model is trying to win a quality benchmark. They’re trying to lower the floor on what it costs to build a media pipeline that actually ships. On that measure, this is the most practical Google AI developer launch of 2026 so far.