Blog

Nano Banana Fashion Reference Guide

Organize Nano Banana fashion edits around an approved identity, product references, controlled variations and a collection review, with no fabricated model scores.

Reference consistencyAbout 6 min read
Editorial visual explaining this topic; not an actual model benchmark.
Illustrative visual created for this article; not a measured model result.

A collection can lose its identity one attractive image at a time. The face becomes a little different, the sweater changes texture and the camera moves closer until the final grid feels like several unrelated shoots. A reference workflow makes those changes visible before they become the collection’s new default. Nano Banana is a family name used across Google image experiences, so begin by recording the exact variant and interface available to you. This article offers a practical fashion editing method, grounded in Google’s reference and iterative-editing guidance. It is not a measured consistency test.

Record the Exact Variant

Displayed variant
Working interface · Acceptance requirement
Approved portrait
Fixed identity cues · Separate styling choices
Person reference
Product reference · Composition reference

The words Nano Banana alone are not a complete production record. Google documents different image variants with different capabilities, and app access is not identical to developer access. Write down the displayed model name, the date, the interface and the available reference controls. That is enough to keep a future repeat from becoming an unexplained comparison between different settings.

Avoid putting a family-wide promise in the brief, such as “Nano Banana always keeps the same person.” Instead, define the actual acceptance requirement: the approved adult model should remain recognizable across the requested campaign images. Store that requirement beside the source portrait. If the interface changes later, you can reassess the workflow without pretending that every image produced under the same nickname came from the same configuration.

Prepare the Identity Reference

Approved portrait

Editorial visual explaining this topic; not an actual model benchmark.
Illustrative visual created for this article; not a measured model result.

Fixed identity cues

Editorial visual explaining this topic; not an actual model benchmark.
Illustrative visual created for this article; not a measured model result.

Two views of the same editorial photograph, not a model comparison.

Choose a clear, approved model image with an unobstructed face and enough context to judge hair, neck and visible proportions. Keep the reference stable throughout the collection. If additional angles are available and you have the necessary permissions, organize them as views of the same identity rather than a collage of vaguely similar people. A reference should settle the casting decision, not reopen it.

Separate identity from styling. Earrings, makeup and hairstyle can change the impression of a face even when its structure seems similar. Decide which features are fixed for this particular collection and which may vary. Then write a short identity note in concrete terms. During review, compare the new candidate with the approved portrait rather than comparing only with the most recent generated image; otherwise small deviations can become hard to notice.

Give the Garment Its Source

  1. Person reference

    Use one approved identity as the casting anchor.

  2. Product reference

    Show the exact sellable variant and relevant details.

  3. Composition reference

    Communicate the desired view without replacing the product facts.

A consistent person wearing an altered product is still an unsuccessful commercial image. Keep the garment reference independent of the portrait and scene. Provide the correct variant, a clear view of the construction and close-ups of important artwork. Where the product has a visible reverse side, a separate back image is more useful than a longer description of details the tool cannot see.

Use reference roles in the request: this image establishes the person, this one establishes the garment, and this one establishes the composition. Google’s image guidance includes multi-image editing examples, but the supported inputs depend on the specific variant and interface. Check the controls you actually have instead of adding references indiscriminately. A smaller, coherent set is easier to review than a folder containing conflicting colorways, poses and faces.

Change One Decision

Choose one variable
Keep the reference anchors · Branch from an approved state
Single-image review
Collection-grid review · Product-detail review
Identity packet
Scene direction · Approved exports

Once the person and garment are settled, choose the next variable deliberately. You might need a different crop for a collection page, a slightly angled pose or another background. Google’s guide recommends using prior images for character consistency and a pose reference for complex poses. Treat those as ways to communicate intent, not as a promise that every detail will survive.

For a useful revision, keep the request narrow enough to diagnose. “Keep the approved person and sweatshirt; show a three-quarter view with hands away from the chest” establishes a clearer review than changing pose, outfit, light, lens and expression together. If the candidate fails, record which element changed. Return to the original garment and identity references for the next branch instead of continually refining a result that already contains an unapproved deviation.

Review the Collection Together

Editorial visual explaining this topic; not an actual model benchmark.
Illustrative visual created for this article; not a measured model result.
  • Single-image review

    Establish that each candidate is individually usable.

  • Collection-grid review

    Look for changes in scale, crop and overall art direction.

  • Product-detail review

    Confirm that the actual garment remains correct.

Review each image first on its own, then assemble the approved candidates at the size they will occupy in the store. The collection view exposes differences that a single-image review can miss: changing head scale, inconsistent crop height, a warmer color cast or a background that no longer belongs to the same art direction. A contact sheet is a comparison aid, not a substitute for detailed garment checks.

Move back to full size for product identity. Compare the neckline, print placement, closures and any small lettering with the source. Make two separate notes: identity or art-direction drift, and product mismatch. Those categories lead to different corrections. Do not approve a wrong garment merely because the face and light match the rest of the series, and do not reject a commercially correct image only because it needs a defined crop adjustment.

Save the Collection Packet

Keep identity, product and scene references separate, then judge their consistency together.

Editorial visual explaining this topic; not an actual model benchmark.
Illustrative visual created for this article; not a measured model result.

Finish with a small packet containing the approved identity, product references, scene direction, export requirements and review decisions. Include the exact variant record, but keep the packet understandable without developer knowledge. Another person should be able to open it and understand what must stay, what may change and which images are already cleared for use.

For teams that prefer a ready visual setup, Studiola organizes Lookbook work around selecting a model when needed, adding your own product and choosing a scene. That can reduce the number of setup decisions between collection items. It does not guarantee identity or product accuracy, so retain the same approval habits. Whether you direct a Nano Banana edit or select a prepared scene, consistency comes from stable references and visible decisions as well as the generation tool.

Sources, method & questions

Published · By Studiola Editorial Team

Frequently asked questions

Does Nano Banana mean one fixed model in every app?

No. Record the exact variant and interface shown in your workflow rather than assuming the family name establishes identical capabilities.

Can I judge collection consistency from one successful image?

One image cannot show consistency across a series. Review multiple actual candidates together and inspect each garment against its own source.

How this guide was researched

An editorial guide based on official sources. No generation benchmark or model ranking was performed. Proposed review steps are a working method, not a measured performance claim.

Sources & further reading

  1. 01
    Google — Nano Banana image generation and reference guidance

    Google · Primary source · Checked

  2. 02
    Google Gemini Apps — generate and edit images

    Google · Primary source · Checked

  3. 03
    Google Merchant Center image requirements

    Google · Primary source · Checked

Prepare a scene with your product

In Studiola, select a model when needed, add your own product and choose a ready Lookbook scene. Review the output against the product reference.

Open Studiola
Nano Banana Fashion Editing: Keep References Consistent