> ## Documentation Index
> Fetch the complete documentation index at: https://docs.yingtu.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Compare Nano Banana models on YingTu

> Choose among Gemini 2.5 Flash Image, Flash Lite Image, Flash Image, and Pro Image using verified price and resolution differences.

Choose a model by the smallest capability set that completes the asset. YingTu
uses one Gemini-native API across the family, so changing models normally means
changing the path ID and validating model-specific size rules.

## Decision table

| Model                         | YingTu sizes     | Current price                     | Best starting point                                |
| ----------------------------- | ---------------- | --------------------------------- | -------------------------------------------------- |
| `gemini-2.5-flash-image`      | Approximately 1K | USD 0.020                         | Existing 2.5 integrations or lowest family price   |
| `gemini-3.1-flash-lite-image` | 1K               | USD 0.025                         | New high-volume work that will remain at 1K        |
| `gemini-3.1-flash-image`      | 1K / 2K / 4K     | USD 0.025 / USD 0.037 / USD 0.055 | General-purpose default and resolution flexibility |
| `gemini-3-pro-image`          | 1K / 2K / 4K     | USD 0.050 / USD 0.050 / USD 0.090 | Complex, premium, or text-sensitive visual work    |

Prices were checked on September 2, 2026. Consult the
[live pricing page](/en/resources/pricing) before production use.

## A practical selection path

1. Keep `gemini-2.5-flash-image` when an existing stable integration works and
   migration has no clear benefit.
2. For a new 1K-only workload, compare Lite with the original model using your
   real prompts; Lite is the current efficiency-focused option.
3. Start most new multi-resolution integrations with
   `gemini-3.1-flash-image`.
4. Test `gemini-3-pro-image` against Flash on the same high-value brief before
   accepting the premium.

## What the live run can and cannot show

The same 1:1 ceramic-object brief completed across all four model IDs. The
outputs confirmed size tiers, edit support, task completion, response fields,
and MIME variation. They do not form a controlled visual-quality benchmark:
the synchronous Flash and Pro requests used higher resolutions than the other
models, and a single prompt cannot represent every production workload.

| Verified observation    | Result                                                              |
| ----------------------- | ------------------------------------------------------------------- |
| 2.5 default and Lite 1K | 1024 × 1024 PNG in this run                                         |
| Flash 2K                | 2048 × 2048 JPEG                                                    |
| Pro 4K                  | 4096 × 4096 JPEG                                                    |
| Asynchronous lifecycle  | HTTP 202, then `in_progress`, then `completed`                      |
| Editing                 | All four preserved the subject and changed the requested background |

<Warning>
  Do not use the observed PNG/JPEG split as a routing rule. File encoding can
  change; `inlineData.mimeType` remains authoritative.
</Warning>

## Open the model guide

<CardGroup cols={2}>
  <Card title="Original Nano Banana" href="/en/models/image/nano-banana/gemini-2.5-flash-image/overview">
    Review migration and fixed-resolution tradeoffs.
  </Card>

  <Card title="Nano Banana 2 Lite" href="/en/models/image/nano-banana/gemini-3.1-flash-lite-image/overview">
    Build a current 1K-only integration.
  </Card>

  <Card title="Nano Banana 2" href="/en/models/image/nano-banana/gemini-3.1-flash-image/overview">
    Use the general-purpose multi-resolution model.
  </Card>

  <Card title="Nano Banana Pro" href="/en/models/image/nano-banana/gemini-3-pro-image/overview">
    Evaluate premium design and 4K work.
  </Card>
</CardGroup>