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

Google released two generative-media options for developers on June 30. The announcement introduced Nano Banana 2 Lite as an image model optimized for throughput and made Gemini Omni Flash available in public preview through Google AI Studio and the Gemini API for video generation and conversational editing. Both also appeared across other Google consumer and enterprise surfaces.

Two media models imply two different queues

Image and video jobs differ sharply in duration, payload size and failure recovery. A product integrating both should not hide them behind one generic generation endpoint. Store the model, prompt, input asset hashes, safety result and output lineage for every job. Use asynchronous status for long video work and define how cancellation affects queued, generating and completed artifacts.

Google’s cost and speed descriptions are product positioning. Teams still need to measure the exact resolution, duration and concurrency mix they expect. An image model suited to high throughput may need different quality gates from a video editor that must maintain continuity across turns.

Build evaluation around edits and provenance

Test initial generation separately from conversational edits. For edits, measure whether requested changes preserve elements the user did not ask to alter. Include repeated revisions, conflicting references and a failed request followed by retry.

Generated media also needs provenance and retention decisions. Keep the relationship between source assets and outputs, surface applicable disclosure metadata, and verify that deletion propagates through previews and cached derivatives. The release broadens API access, but production readiness depends on a media pipeline that can explain, reproduce and retire each generated asset.