Nano Banana Rumor Update: Flow Report Points to 2.1

September 28 update: the reported name has changed

Alexey Shabanov’s September 27 TestingCatalog article reports that the Flow web build replaced the earlier Nano Banana 2.5 Flash label with Nano Banana 2.1. We verified the public report, but did not reproduce or authenticate the build change. TestingCatalog follow-up

This is a follow-up from the same outlet associated with the earlier sighting, not independent corroboration. It leaves the eventual product name unresolved. Our original September 27 reporting appears below; references there to 2.5 Flash describe the earlier claim. No confirmed profile or new model route has been created from this reported label. TestingCatalog follow-up

A smaller version increment is not a measurement of the underlying change. It cannot establish whether an eventual update improves editing, preserves a subject more reliably, or changes the effort needed to obtain an acceptable image. Treat the name as a tracking clue and keep the evaluation criteria unchanged. For an existing image project, retain the working model, references and acceptance checks until a usable, documented alternative can be evaluated. TestingCatalog follow-up

Rumor evidence: what each signal can establish

Where the Nano Banana 2.5 Flash rumor starts

The accessible reporting trail includes a September 5 TestingCatalog article describing code references to Nano Banana 2.5 Flash. Its embedded social post is dated September 4. That establishes a public report about the name, not a Google announcement. Evidence context.

The later Flow lead is narrower. A September 24 PromptBluePrints report describes a screenshot attributed to TestingCatalog, showing the name in a character-creation model selector. The article links the originating X post. Our direct attempt to retrieve that post returned a forbidden response, so this coverage relies on the accessible secondary account for the Flow claim. We did not independently authenticate the screenshot or reproduce the interface. Evidence context.

The two dated reports should not be counted as two independent confirmations of a working model. They are associated with the same reporting outlet, and a label appearing in different contexts may still refer to a single experiment. The useful development is the reported move from code references toward an application surface. Whether that movement reflects a release candidate, a limited test or an unused option remains unresolved. Evidence context.

Why a Flow appearance would matter for image creators

If the reported option becomes usable, its significance would be where image creation happens. A model embedded in an existing creative workflow could influence which system people use for character references, scene concepts and revisions before they assemble a video. Evidence context.

That is our interpretation of the potential impact, not a capability claim about Nano Banana 2.5 Flash. A character-related interface does not demonstrate character consistency, multi-reference editing or control over identity. Those abilities would need their own documentation and output evidence. Similarly, presence inside an application would not establish an API offering, an unrestricted rollout or equivalent behavior across products. Evidence context.

For a creator, the consequential questions are concrete: can the model preserve a chosen subject across revisions, follow a layout instruction and return usable files under known limits? For a developer, the questions concern a stable identifier, reproducible settings and a documented access route. A rumor about placement can motivate those questions, but cannot answer them. The impact could be substantial if a dependable model arrives at a convenient point in the workflow; convenience alone would not establish better images. Evidence context.

Spicy Mayo is a separate identity question

Community reports associate the anonymous name Spicy Mayo with a possible Nano Banana successor. LLMLearner records that association as a rumor, tracing discussion to September 9 social posts. We have not established that Spicy Mayo and the reported Flow entry are the same system. Evidence context.

A codename, a public-facing label and an API model identifier are different pieces of evidence. Joining them into one identity requires a disclosure or another independently checkable link. Similar timing, naming patterns and enthusiastic examples may make a connection seem plausible, but they do not supply that link. We therefore do not transfer any reported results from Spicy Mayo to Nano Banana 2.5 Flash. Evidence context.

This distinction also prevents accidental specification laundering. An informal image comparison can concern a transient test build, while a later product uses a different checkpoint or settings. Even a real test output would not automatically describe the eventual product. Readers should preserve the source label when discussing an example rather than silently renaming it. Until the identity is established, the Flow rumor and the Arena-related identity speculation remain parallel leads with different evidentiary limits. Evidence context.

What the official directory settles and leaves open

Google’s public Gemini API directory is useful for checking documented offerings. At our September 27 inspection it listed Nano Banana 2, Nano Banana 2 Lite and Nano Banana Pro. We found no entry using the Nano Banana 2.5 Flash name. Evidence context.

This observation supports a limited conclusion: the directory we checked did not provide a public model contract for that name. It does not exclude private partner testing, a future addition or an app-only feature. Conversely, the existence of documented neighboring models does not let us fill the gaps by copying their limits. We have not established supported inputs, output resolution, reference count, editing controls, downloadable weights, API availability or cost for the rumored model. Evidence context.

The digits deserve particular care. A familiar family name can make an unannounced version sound like a routine upgrade with inherited capabilities. That is not a reliable basis for procurement or a delivery promise. Any eventual documentation should be read as a new contract: identify the exact model, check which product exposes it and separate previews from broad access. An announcement, a selectable app option and a callable API can each happen at different times. Evidence context.

What to do before changing an image-model workflow

Keep a working production route while preparing a small evaluation brief. There is no verified release date in this review, so a project deadline should not depend on the rumored model arriving. Waiting can carry a real opportunity cost even when the eventual release turns out to be useful. Evidence context.

Write down a few representative image tasks before seeing promotional examples: a product reference that must preserve its markings, a composition with specified object placement, and an edit that should leave the surrounding scene unchanged. These are proposed checks, not tests we performed on the rumored model. Keep the original prompts and references so an eventual comparison answers your actual requirements instead of a task selected to flatter a particular output. Evidence context.

Once access is documented, record the exact version, settings, number of attempts and reasons for rejecting outputs. Review images at a matched scale, and separate visual preference from instruction failures. The existing image review guide and downloadable worksheet can help organize that record. Do not label existing demonstration assets as Nano Banana 2.5 samples: this article contains none, and the site’s existing examples are not evidence about this rumor. Evidence context.

What would change this report

The next meaningful update would add evidence, not simply another repetition of the name. An official image-model page, a documented public access route or an attributable provider clarification would each resolve a different part of the story. Evidence context.

A model card could clarify identity and supported operations. A documented access route could clarify who can use it and under what conditions. Reproducible, versioned output comparisons could help readers evaluate practical behavior once access exists. None of these should be inferred from the others: a product announcement is not a quality measurement, and a strong sample is not an availability notice. Evidence context.

We will revisit this report when the evidence changes, including if the label disappears, the identity is corrected or the expected product never materializes. The original publication date will remain visible, with dated corrections describing what changed. For now, this is a high-impact possibility with limited public evidence. Its relevance comes from the potential effect on image creation and editing, while confidence in identity, timing and capabilities remains separately unresolved. Evidence context.

FAQ