LlamaGen logo

Nano Banana 2.1 workflows

Edit an image with Nano Banana 2.1 without losing the brief

A useful edit starts with a clear distinction between the intended change and the details you need to preserve. Describe both, then inspect the returned image for changes you did not request.

What you will make

An edited original image with a written preservation checklist and a clear next revision.

Create with Nano Banana 2.1

Nano Banana 2.1 requires a paid LlamaGen plan. Check the current credit charge and export choices before generating. Direct Google Cloud API billing is separate. Plans and credits

A finished AI-generated visual presented in an everyday creative setting.
AI presentation concept: a finished visual shown in use. This scene is separate from the direct model output below.

The hero is an AI-generated presentation concept showing a finished result in context. It is not a customer photograph, application screenshot or a separate test of the model’s performance.

LlamaGen editorial resources

Content reviewed:

Inspect a generated example

The example below is a direct model output from our documented generation workflow. Review the actual image, rather than assuming the model followed every instruction.

Reference images used in the generation workflow

These are owned, fictional reference assets. Their inclusion does not guarantee exact identity or product detail in the output.

Direct Nano Banana 2.1 image-generation output for the workflow described on this page.
A direct Nano Banana 2.1 output from the documented workflow. Inspect the image for the checks described in this guide.

Write the change and the preservation rules

Choose one owned input image and one main change. For a story image, that might be a different environment or lighting treatment while preserving the original adult character’s identifying features and key prop.

Avoid mixing a background replacement, a new pose, a costume redesign and a new camera angle in an unclear first request. A focused brief makes unwanted changes easier to recognize, even though it cannot guarantee perfect subject preservation.

  • State the intended change in one sentence.
  • List the face, costume, prop or composition details to keep.
  • Check that you have permission to edit the source image.

Inspect more than the changed area

Look at the intended edit first, then compare the rest of the image with the reference. A convincing new background is not enough if a hand, face or important object changed in a way that breaks the intended use.

Keep the original and useful intermediate versions. If the result drifts, revise the brief using the original reference rather than repeatedly editing an image whose important details have already changed.

  • Check facial features, hands and subject proportions.
  • Check object shape, colors and identifying details.
  • Inspect edges and the transition between subject and scene.

Treat conversational editing as a review loop

Google documents multi-turn image editing, but its model card also notes that character consistency and instruction following can be imperfect. A later edit may introduce new deviations.

Describe a concrete correction and inspect its side effects. Do not call a flat image an editable layered source, or imply that prompt-based editing creates independently adjustable text, borders and subjects.

  • Preserve the brief alongside each kept version.
  • Separate an image edit from rebuilding a file with layers.
  • Use a dedicated editor when a precise manual correction is the better finishing step.

A preservation checklist

Use a small list that fits your actual image instead of asking the model to keep everything unchanged.

ElementDescribe what to keepInspect in the result
Character identityIdentifying face, hairstyle and adult character designUnwanted changes to features or proportions
Costume and propDistinctive clothing and the key story objectNew accessories, missing details or altered shape
Intended editThe one environment, lighting or visual changeWhether it happened clearly and coherently
Output useFinal crop and space for later textClipped subjects, unreadable details or unwanted lettering

A practical workflow

A focused edit brief

Use your own original character image. The aim is to make the intended change easy to review while treating identity preservation as a check, not a guarantee.

A starting prompt to adapt

Edit the supplied original character image. Change the environment and lighting to a quiet evening scene. Keep the character’s identifying features, clothing and key prop recognizable. Do not add new people, a logo or written text.

This brief is an adaptable recipe, not the verbatim request log for the pictured output. It does not promise identical results or a successful first generation. The generation record preserves the original request.

Try the workflow

  1. Save the original reference and write the preservation list.
  2. Generate a candidate edit with a focused change request.
  3. Compare original and result before deciding on the next correction.

What to inspect

  • Was the intended change actually made?
  • Did the face, hands or key prop drift?
  • Would the result work at its final crop and display size?

Questions before you begin

Will Nano Banana 2.1 keep every part of the source unchanged?

No. Reference editing can change details you did not request. Google’s model card lists imperfect consistency and instruction following among the limitations, so inspect the entire result.

Can I edit the same image several times?

The model supports multi-turn editing, but later turns still need review. Keep the original and useful intermediate files so you can recover from drift.

Does this create editable layers?

A generated or edited image is not automatically a layered design file. Keep source materials and use an editor suited to precise layer, text or layout work when needed.