Google made two generative media models available to developers via the Gemini API today: Nano Banana 2 Lite, a fast image generator that produces a 1K image in about four seconds at $0.034 a shot, and Gemini Omni Flash, a conversational video model that animates those images into 10-second clips for $0.10 per second. Both run through the same google-genai
SDK. Together, they form the cheapest complete image-to-video pipeline in the Gemini ecosystem — roughly $1.04 per finished clip. If you have been waiting for a cost-effective API path from text prompt to animated video, this is the stack.
Nano Banana 2 Lite: The Fast Image Model #
Nano Banana 2 Lite carries the model ID gemini-3.1-flash-lite-image
and is the newest member of Google’s Nano Banana family — the umbrella name for Gemini image models. It generates images at 1K resolution across 14 aspect ratios, applying an invisible SynthID watermark automatically. Speed in practice lands at four to six seconds, which makes it well suited to pipelines where you generate dozens of variants and pick the best few.
The pricing is the real headline: $0.034 per image at standard rates, or $0.0168 under batch pricing. That is roughly half the cost of the standard Nano Banana 2 model ($0.067 per image). Despite the “Lite” label, its Text-to-Image Arena Elo of 1,251 actually beats Nano Banana Pro at 1,245, which suggests you are not trading much quality for the lower price at standard prompt complexity.
There are limits worth stating plainly. Nano Banana 2 Lite caps at 1K resolution — the standard Nano Banana 2 goes up to 4K. It also struggles with highly complex multi-object prompts that require precise spatial arrangement. If your prompt has five specific objects in specific positions, use the standard model. If you are generating thumbnail variants, ad drafts, storyboard frames, or UI mockups at scale, Lite is the better choice.
Gemini Omni Flash: Conversational Video Generation #
Gemini Omni Flash (gemini-omni-flash-preview
) is a natively multimodal video model that accepts text, image, and video inputs in combination. At launch, it produces 10-second clips at 720p for $0.10 per second. Google has signaled that longer durations are on the roadmap.
The feature that distinguishes Omni Flash from Veo 3.1 Fast — which carries the same $0.10/sec price — is conversational editing. You generate a clip, then send a follow-up instruction in the same session: “make the camera movement slower” or “change the lighting to dusk.” The model maintains context across turns through the Interactions API. This is not a stateless generation call, and that difference matters if you are building tools for iterative creative work.
Be clear on what Omni Flash does not do. It does not support Computer Use, image generation, audio generation, or the Live API. It is paid-tier only. As a preview model, it can introduce breaking changes in minor releases — plan for that in production environments.
The Image-to-Video Pipeline #
The intended developer workflow is to chain the two models. Generate a still with Nano Banana 2 Lite, then pass that image to Omni Flash to animate it. Both calls use generate_content()
through the same client, which keeps the plumbing straightforward:
from google import genai
from google.genai import types
client = genai.Client(api_key="YOUR_GEMINI_API_KEY")
img_response = client.models.generate_content(
model="gemini-3.1-flash-lite-image",
contents="A futuristic cityscape at dawn, neon-lit skyscrapers",
)
image_part = next(
p for p in img_response.candidates[0].content.parts if p.inline_data
)
vid_response = client.models.generate_content(
model="gemini-omni-flash-preview",
contents=[
types.Part(inline_data=image_part.inline_data),
"Camera slowly pans right, clouds drift, lights flicker on",
],
)
Total cost for a generated image plus a 10-second clip: approximately $1.034. Google has published three remixable demo applications in Google AI Studio to illustrate the pipeline: Anywhere (landmark selfies animated), Space Lift (interior design concepts from room photos), and Omni Product Studio (static product images to e-commerce video). They are worth inspecting before you build your own implementation.
Imagen 4 Users: This Is Your Migration Path #
If you are still running Imagen 4, the August 17 shutdown deadline makes this evaluation unavoidable. The migration carries three breaking changes: generate_images()
is gone, images now arrive as content parts rather than a dedicated image response object, and the number_of_images
parameter does not exist on Nano Banana models. For most teams rebuilding that pipeline, Nano Banana 2 Lite is a reasonable landing spot for volume workloads, while standard gemini-3.1-flash-image
handles cases where quality ceiling matters more than throughput cost.
Get Started #
Both models are available now in Google AI Studio, the Gemini API, and the Gemini Enterprise Agent Platform. Nano Banana 2 Lite is available on the free tier with rate limits; Gemini Omni Flash is paid-tier only. Use the current google-genai
SDK — not the older google-generativeai
package. Full pricing is on the Gemini API pricing page. Google’s official launch post includes additional context on the three demo apps and the broader generative media roadmap.