Compare Nano Banana image models

Start with the smallest current model that can complete the asset and remain maintainable for the integration’s expected lifetime. The request shape is shared, but size rules, lifecycle, and intended workload differ.

Current decision table

Model ID	YingTu output tier	Lifecycle	Relative cost position	Start here when
`gemini-2.5-flash-image`	Approximately 1K; omit `imageSize`	Google lists earliest shutdown on 2026-10-02	Lowest current family price	Maintaining a legacy integration while actively migrating
`gemini-3.1-flash-lite-image`	1K only	Current stable model	Same current 1K price as Flash	A new job will remain at 1K and efficiency is the priority
`gemini-3.1-flash-image`	1K, 2K, 4K	Current stable model	Mainstream	A new integration needs a general default or multiple resolutions
`gemini-3-pro-image`	1K, 2K, 4K	Current stable model	Premium	Complex layouts, text-sensitive graphics, or high-value assets justify comparison

Exact prices change more often than this decision logic. Use the dated YingTu pricing page before estimating production.

Selection path

For a new 1K-only workload, compare Lite with Flash using representative prompts and acceptance criteria.
Start with Flash when the same integration may need 2K, 4K, or extended aspect ratios.
Compare Pro against Flash only on work whose value or complexity can justify the premium and larger responses.
Treat 2.5 as a migration source, not the default for a new long-lived integration.

Migrating from Gemini 2.5 Flash Image

Google’s image guide recommends moving legacy users toward Nano Banana 2 Lite. On YingTu:

choose Lite when 1K remains sufficient;
choose Flash when the integration needs higher resolutions or the broader general-purpose path;
remove assumptions that the response is always PNG or always starts with an image part;
add imageSize: "1K" only after switching away from 2.5;
retest prompts, source images, output acceptance, timeouts, and cost.

Compare quality with your own acceptance criteria

Model names and provider examples cannot identify a universal quality winner. Compare the models with the same prompts, source images, output size, and review criteria. Track accepted outputs, retries, manual correction time, latency, and total request cost before moving production traffic.

Review Lite

Choose a current 1K-only path.

Review Flash

Build a general multi-resolution integration.

Review Pro

Plan premium and 4K work.

Plan the 2.5 migration

Preserve a legacy path without ignoring the retirement risk.

​Current decision table

​Selection path

​Migrating from Gemini 2.5 Flash Image

​Compare quality with your own acceptance criteria