ARTICLE

29 Oct 2026 Nano Banana 2 (`gemini-3.1-flash-image`) shutdown: cut over Bubble bulk routes to Nano Banana 2.1 in BIAIG

A Bubble-focused cutover checklist: inventory model IDs in fields and API Connector, route to `gemini-nano-banana-2.1`, keep BIAIG catalog rollback rows—while noting Google’s deprecations page (7 Oct 2026) still lists no firm shutdown date despite digest/third-party 29 Oct claims.

Why now

Editorial digests keep citing a reported shutdown of gemini-3.1-flash-image (Nano Banana 2) on 29 October 2026. On 6 October, Google shipped gemini-nano-banana-2.1 (Nano Banana 2.1) as GA—the named successor. If your Bubble bulk routes still hit the old model ID through API Connector or store it on data types, I would not wait for a formal letter. You need a dated cutover in internal BIAIG routing. I am not reopening third-provider strategy here; that is covered elsewhere. Worth reading next: Imagen 3 GA endpoints discontinued:. Worth reading next: Imagen 3 GA endpoints discontinued:.

Why now

On the date—I stay precise. Google’s primary lifecycle docs (release notes / deprecations, last updated 7 Oct 2026 UTC) mark gemini-3.1-flash-image deprecated with replacement gemini-nano-banana-2.1, but shutdown is “No shutdown date announced”, with separate wording that exact retirements get advance notice. Third-party aggregators (e.g. AI Change Watch) cite 2026-10-29. I treat that as unverified until the same date shows up on ai.google.dev. What I do as a builder: plan cutover toward 29 Oct as an operational deadline, but keep rollback until Google publishes a firm shutdown.


Target state (one line)

Primary routing: gemini-3.1-flash-image → gemini-nano-banana-2.1. Retired preview IDs such as gemini-3.1-flash-image-preview (Google: shutdown 25 Jun 2026) should not stay in production or rollback “just in case.”

BIAIG is the internal catalog/routing pattern from earlier Gemini/Nano Banana pieces—not a Google or Bubble product name in public documentation. A closer look: SenseNova U1 Pro and 8K. A closer look: SenseNova U1 Pro and 8K.


Step 1 — Bubble inventory

  1. Data fields / option sets / catalog tables: stored model_id, route keys, prompt templates embedding model strings.
  2. API Connector: every call whose URL, body, or headers still send gemini-3.1-flash-image.
  3. Backend workflows: scheduled and bulk retries often duplicate model IDs outside Connector definitions.
  4. Batch vs sync: Batch API jobs (~50% of standard cost, ~24h target turnaround, generateContent-style batches) need their own list—cutover testing requires polling, not a synchronous Connector assumption.
  5. Preview leftovers: remove from production routing immediately.

Step 1 — Bubble inventory

Deliverable: workflow → field/connector → current model ID → sync or batch → criticality.


Step 2 — rerun-breaking deltas after the swap

Topic gemini-3.1-flash-image gemini-nano-banana-2.1 Bubble action
Resolution 0.5K / 512px supported (image-generation guide) 1K / 2K / 4K only Catalog rows on 512px need a new size or downstream resize—a model swap alone fails
Thinking default minimal medium on 2.1 Re-check latency and cost; image models may bill thinking/interim tokens
Reference images different limits up to 14 refs, up to 4 characters / 10 objects (model card) High-ref bulk jobs may degrade silently
Video → image older changelog text sounded 3.1 Flash Image–only current guide lists 2.1, 3.1 Flash, 3.1 Flash Lite Plan from the current image-generation guide
Batch API supported supported Keep batch for non-urgent bulk; store job IDs and poll

Step 2 — rerun-breaking deltas after the swap


Step 3 — stage in BIAIG with rollback

  1. Add a primary catalog row for gemini-nano-banana-2.1 with updated defaults (resolution, thinking level).
  2. Keep a rollback row for gemini-3.1-flash-image disabled in production, enabled for staging/canary until Google confirms shutdown.
  3. Canary one low-risk bulk workflow; compare output dimensions, failures, time-to-ready (sync or batch completion).
  4. Re-benchmark cost: pricing fetched 8 Oct 2026 still shows paid/batch rates under “Gemini 3.1 Flash Image (Nano Banana 2)” with no separate Nano Banana 2.1 section—do not assume price parity without a published 2.1 row; measure on real payloads.
  5. Production flip: update BIAIG routing and Connector defaults; do not delete the old catalog row before an official shutdown date.

Step 4 — before 29 Oct 2026 (or any new date on ai.google.dev)

  • Re-read Gemini deprecations and changelog—not digest-only sources.
  • Production traffic on gemini-nano-banana-2.1; zero catalog rows on preview IDs.
  • 0.5K or >14-ref jobs migrated or explicitly excluded from the 2.1 route.
  • Batch jobs: scheduler + persisted batch_job_id; UI not blocked on sync waits.
  • Runbook: rollback = re-enable old catalog row + revert Connector default.

Source bounds

Official: Gemini API release notes (6 Oct 2026 GA for 2.1), deprecations (7 Oct 2026), model cards, image-generation guide, Batch API, pricing (8 Oct 2026). Digests motivate planning; 29 Oct is an operational hypothesis while Google states no shutdown date announced.

Internal methodology draft—not for external publication.

← All articles