Skip to main content

Nano Banana 2 Lite

The efficiency specialist of the Nano Banana family — the same generation and editing surface as Nano Banana 2, pinned to 1K output for sub-2-second latency at roughly half the cost.

Capabilities​

FeatureSupport
Text-to-ImageYes
Image-to-Image (editing)Yes
Resolution1K (1024px) only
Reference ImagesUp to 14 (10 objects + 4 characters)
Aspect Ratios1:1, 1:4, 1:8, 2:3, 3:2, 3:4, 4:1, 4:3, 4:5, 5:4, 8:1, 9:16, 16:9, 21:9
Thinking LevelMinimal (default) or High
SeedYes
Negative PromptNo
Inpainting / MaskNo

When to use Lite vs. Nano Banana 2​

You needPick
Sub-2s turnaround for rapid-fire iterationLite
Many cheap variants of the same ideaLite
2K / 4K output for deliveryNano Banana 2
0.5K draft thumbnailsNano Banana 2

Prompts, reference-image behavior, and aspect handling are identical across the two — you can iterate on Lite and re-run the exact same node on Nano Banana 2 for the final render.

Thinking Level​

Lite exposes a two-step thinking control:

  • Minimal (default) — fastest path; the model renders directly from the prompt.
  • High — the model reasons about composition before rendering. Slower, but noticeably better on prompts with several constraints (layout + text + multiple subjects).

Prompting Tips​

  • Treat it as your sketchpad. At ~2 seconds per image, generating 10 variants costs less time than perfecting one prompt. Explore wide, then upgrade the winner to Nano Banana 2 at 4K.
  • Reference images work the same as Nano Banana 2 — up to 14, with @-mentions for per-image roles. See Referencing Images in Prompts.
  • Bump Thinking to High for layout-heavy prompts. Infographics, posters, and multi-subject scenes benefit the most.

Limitations​

  • 1K output only — no 0.5K draft tier and no 2K/4K delivery tiers
  • No negative prompt, no inpainting
  • Preview model — API may change

See Also​