As reviewed on 28 October 2026, ChatCut’s pinned nano-banana selector means Nano Banana Pro, gemini-3-pro-image. Nano Banana 2 is the separate gemini-3.1-flash-image: it produced a small image in a direct development check but has no selector in the inspected integration. Pro documents 1K/2K/4K and up to 14 references. Full-boundary generation and deployed production remain unverified, and the saved Pro response exposes a final-image selection bug.
Product names and route evidence
Original Nano Banana is gemini-2.5-flash-image; it is neither Pro nor Nano Banana 2. Do not infer a model from the family nickname alone. The table distinguishes native documentation, ChatCut’s inspected controls and the exact development checks.
| Model and public ID | Capabilities, integration and evidence |
|---|---|
Nano Banana Progemini-3-pro-imageAvailability: Pinned integration and saved dev checks; production unverified. Source links describe native specifications. Reviewed: | Native documentation: Text/image generation and edits; 1K/2K/4K tiers; up to 14 image references documented. ChatCut integration: nano-banana selects Pro; local reference cap 14 and 1K/2K/4K controls exposed. Endpoint evidence: Final 1024×1024 PNG succeeded at 1K. 2K/4K and full-reference boundaries unverified. Saved adapter replay selects an intermediate thought image. |
Nano Banana 2gemini-3.1-flash-imageAvailability: Pinned integration and saved dev checks; production unverified. Source links describe native specifications. Reviewed: | Native documentation: Distinct model; documented 512/1K/2K/4K tiers and up to 14 images. ChatCut integration: No separate selector in pinned ChatCut integration; nano-banana continues to select Pro. Endpoint evidence: Final 512×512 PNG succeeded in a direct development check. This does not establish a ChatCut selector or full-boundary generation. |
Which sizes and references are documented?
Pro documents 1K, 2K and 4K settings. These are native tiers, not universally experimental options. The Google 16:9 table lists 1376×768, 2752×1536 and 5504×3072 respectively; dimensions vary by aspect ratio. The saved endpoint success used 1K at 1:1, returning 1024×1024. Empty-content and malformed-image probes at 2K/4K did not establish either successful output or resolution rejection. The inspected ChatCut tool’s tier list is not proof of completed generation at those tiers. Pro model card.
Pro’s documented maximum is 14 image references, and ChatCut enforces a local 14-reference cap. Google’s role guidance allows up to 6 object, 5 character and 3 style images within that total; those are maxima, not a promise to preserve 14 distinct identities. Token-count requests accepted malformed images and therefore did not test generation at the 14/15 boundary. Use a few clear references first, name each reference’s purpose and inspect the result.
The Pro card documents PNG/JPEG/WebP and HEIC/HEIF, with a 7 MB inline-image limit. ChatCut’s public-URL path has a separate 5 MiB limit and HEIF normalization. Do not assume every project format is converted; rasterize SVG before submission. Native transport limits do not override a stricter attachment path. See generation references.
How does Nano Banana 2 differ?
Nano Banana 2 documents 512, 1K, 2K and 4K, up to 14 image references, and additional aspect ratios such as 1:4, 4:1, 1:8 and 8:1. Its native model card describes image/video inputs; that does not add a video input to ChatCut’s still-image tool. The direct development check returned a 512×512 final PNG for the exact model. This does not create a selector in the pinned integration. Nano Banana 2 model card.
There is no matched Pro-versus-2 quality, speed or cost benchmark here. For a product sheet, first verify the model actually shown in your app, then use one product reference, a requested ratio and a small test. Compare silhouette, logo/text accuracy and composition manually. Higher resolution and more references do not establish better identity consistency.
What has actually been tested?
Pro access is proven by a successful response containing both an intermediate thought image and a final non-thought image. A saved replay of the pinned ChatCut adapter selected the first image, which was the thought image. HTTP 200 therefore does not prove that ChatCut delivered the correct final image. No Desktop import run in that check established correct delivery. Inspect the asset before using it; if an intermediate or unexpected image appears, report it through workflow failures with the model and settings.
The checked tool does not expose the full native conversational session, search-grounding configuration or a transparent-alpha control. These models generate still images; audio generation and timed video output are not still-image capabilities. Provider documentation about thinking or editing does not prove every native control is available in ChatCut.
Make a product-sheet draft
Confirm the selector
Check the model available in your app. In the inspected integration, nano-banana means Pro. Do not request Nano Banana 2 through that alias.
Use one clear reference
Attach a product photo and ask for one 1K image at the desired ratio: keep the silhouette and colors; show front and side views on plain gray; do not add text. State which details may change.
Review before assembly
Check the quoted model/settings before submitting. Inspect the delivered image for a final result, product accuracy and actual dimensions, then place it from My Assets on the timeline for crop, scale and duration edits.
This produces a raster asset, not individually editable product views or vector artwork. Continue revisions by referencing the selected result. The image-generation guide explains the shared task flow; GPT Image has different reference and background rules.
How should I compare costs?
No matching current Pro production quote was verified. The older shared Nano Banana card estimate names Nano Banana 2, whereas the pinned selector maps to Pro, so its number is not a Pro price. Native Google billing is token-based and varies by service and settings; it is not a ChatCut credit conversion.
Use the current preflight estimate with the exact model ID, account catalog, requested tier, actual dimensions when known, ratio, quality and image count. Compare on the same quote date and inspect Credits history → Detail after completion. Agent usage is separate. Credits Policy is the current rate owner; generation credits explains estimated versus settled charges.
Frequently asked questions
What does nano-banana select in ChatCut?
Can I select Nano Banana 2 in the pinned build?
Does Pro support 2K and 4K?
Did the 14-reference boundary pass?
Does a successful Pro response guarantee the final image was delivered?
Is Pro better or cheaper than Nano Banana 2 for product sheets?
Last verified: 28 October 2026. Public sources, accessed 27 October: Google image guide, Pro model card and Nano Banana 2 model card. The pinned integration, saved endpoint results and adapter replay were reviewed on 28 October. Production, full-reference delivery and higher-tier completion remain unverified; no new paid generation was performed. See the model matrix for other model families.