FlowtandBriefAI · Software · ResearchMeet Flowtand

Google opens Nano Banana 2 Lite and Gemini Omni Flash to developers

Announcement

Google DeepMind

  • Models
  • #gemini
  • #nano banana
  • #video generation

Browse Models topic page →

Google DeepMind says Nano Banana 2 Lite, a fast low-cost image model, is available in AI Studio, the Gemini API and Gemini Enterprise Agent Platform, and that Gemini Omni Flash, a preview video generation and conversational editing model, is reaching developers for the first time. Google positions the pair as a chained image-to-video workflow and lists several documented limitations for the video model.

What Google announced

Google DeepMind product managers Alisa Fortin and Anish Nangia announced two models. Nano Banana 2 Lite (gemini-3.1-flash-lite-image) is described as the fastest and most cost-efficient image model in the family, aimed at rapid ideation and high-volume pipelines, and Google recommends it as a replacement for the first Nano Banana (gemini-2.5-flash-image). It is listed in AI Studio, the Gemini API and Gemini Enterprise Agent Platform, and also rolling out to consumer surfaces. Gemini Omni Flash (gemini-omni-flash-preview) supports video generation and conversational editing from text, image and video inputs; Google says it is priced at $0.10 per second of output, matching Veo 3.1 Fast. Google reports a 4-second latency and $0.034 per 1K-resolution image for the Lite model. Omni Flash is a public preview with stated limits: 10-second generations, no audio reference upload or scene extension in the API, 3-second video references accepted by the schema but not correctly processed, and limited character consistency across scene changes. Google suggests chaining the two: generate an image, pass it as a reference for animation, and use the Interactions API to keep session context for up to three sequential edits.

A plausible application for a retail content producer

A retail content producer who needs product visuals for many SKUs before a campaign could try a two-stage pipeline: draft a product image with the Lite model, then animate an approved draft into a short clip with Omni Flash. This maps to the demo pattern Google describes for converting static images into e-commerce videos. Because Omni Flash is a preview with 10-second outputs and limited scene-change consistency, the producer should scope the experiment to single-scene, short-form assets rather than narrative spots, and should treat generated text rendering as something to verify visually before use.

How to trial it against a baseline

A team could run a small paired test: generate a set of images on the model already in use and the same set on Nano Banana 2 Lite, then measure generation time and how often prompts are followed, judged by a human reviewer. For video, animate only images that pass review and check whether the clip keeps the subject consistent. Record failure cases explicitly: garbled in-image text, drift between scenes, or clips that hit the 10-second ceiling. Because the vendor's latency and cost figures are self-reported, compare them against your own logged timings. Keep a human approval step before any generated asset reaches customers, and note that Omni Flash output should be checked for correctness since it draws on general knowledge.

Trade-offs and when to hold back

Choosing the Lite model trades image quality and control for speed and cost; Google positions its generalist and Pro models for cases where accuracy matters more. Omni Flash carries preview risk: undocumented changes are likely, several features are missing, and the vendor states scene-change consistency is imperfect. Prerequisites include access to the Gemini API or AI Studio and a workflow that can store and pass generated images between model calls. Teams should not use this pairing where clips must exceed 10 seconds, where audio references or scene extension are essential, or where a stable, fully supported model is contractually required. Whether the recommended swap from the first Nano Banana model is seamless should be confirmed against the developer documentation.

Context and limitations

  • Based on a short source excerpt; the full article was not analyzed.
  • Performance, latency and cost figures are Google's own claims, not independent test results.
  • Omni Flash is a public preview with documented gaps, including 10-second output and unprocessed 3-second video references in the API.
  • Based on bounded source text, not independent testing. Practical suggestions are editorial proposals.