Visual PromptCompiler / image editionLocal-only V1
N2
Model adapter / v1.0.0 / noindex until live QA

Nano Banana 2 prompt adapter

Compile a visual spec into conversational reference intent, an image response plan, explicit text guidance, and current Gemini 3.1 Flash Image limitations.

Official model ID
gemini-3.1-flash-image
Source checked
July 28, 2026
Model card updated
July 21, 2026
BriefintentSpeclocksAdaptermodelOutputready
Working surface / no account

Compile for Nano Banana 2

The adapter explains what each reference should control, favors conversational edit language, and separates planning JSON from unsupported API structured output.

Stays hereSaved in this browser only
Task preset
SPEC / 01
01 Outcome
02 Lock the subject
03 Direct the scene
04 Set change boundaries
05 Assign reference roles

Describe references only. Images are never uploaded or stored here.

06 Choose the target
Workflow
Ready to compile locally

RULES / 06

What this adapter changes

01

Conversational references

References are described by intent instead of being copied into Seedream's ordered Image 1…N syntax.

02

Preserve unmentioned regions

Edit mode tells the model what to change, then explicitly protects every visual property that was not named.

03

Resolution and ratio plan

The export uses Gemini's image response fields with an explicit aspect ratio and a 2K planning choice.

04

Text workflow warning

When visible text is requested, the result advises drafting exact wording first and then generating the image with that text.

05

Fidelity boundaries

Character tasks disclose the documented four-character resemblance boundary, and every result discloses the ten-object-reference-image fidelity boundary.

06

API truthfulness

The JSON tab is labeled as portable planning data because this image model does not support Gemini structured outputs.

Official source boundary

Current Gemini naming, kept separate

Nano Banana 2 is the official name for Gemini 3.1 Flash Image, model code gemini-3.1-flash-image. The adapter does not mix in Lite or Pro-only claims. It records SynthID, output-count variability, provider-side resolution choices, and conversational editing from Google's image generation documentation. This V1 keeps the planning export fixed at 2K.

Image generation guide ↗Model card ↗
QA / GATE

Six live model tasks still required

The adapter contract is locally tested, but no authenticated Google AI Studio run has been performed from this workspace. These rows are a required test plan, not invented results. The page remainsnoindex and excluded from the sitemap until all six tasks have been executed, at least four pass, and the passing set includes one Character and one Product test.

IDTaskPrimary checkState
N2-01Character portraitIdentity, face, outfitPending live run
N2-02Same character, new sceneDrift and reference intentPending live run
N2-03Product heroShape, material, labelPending live run
N2-04Product lifestyleProduct lock vs. scenePending live run
N2-05Text-heavy posterText and hierarchyPending live run
N2-06Conversational editUnchanged region fidelityPending live run
SUITED

Good reasons to use this adapter

  • You want fast general image generation and conversational edits.
  • You need reliable text-oriented image guidance.
  • You need several reference intents in one readable brief.
  • You want this V1's fixed 2K planning export.
LIMITS

Disclosures the prompt must carry

  • Generated images include SynthID.
  • The requested number of outputs may vary.
  • Exact typography still needs human inspection.
  • Planning JSON is not API structured output.
Use the adapter inside a task

Start from the work, not the model name.

01Character consistency
Identity source and scene change02Product photography
Product source and campaign direction

Visual Prompt Compiler is an independent tool and is not affiliated with Google. Gemini and Nano Banana model names belong to Google and their respective owners.