Verified against Nano Banana / Gemini 3.1 Flash Image · 2026-07-21
Make one surgical edit to a photo without touching anything else
A conversational edit instruction for an uploaded photo that isolates exactly one change and explicitly locks everything else — lighting, pose, background, composition — so Nano Banana's edit doesn't quietly redraw the whole frame.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are editing the photo I've uploaded in this conversation. This is a single, targeted edit — not a regeneration. Everything not named in the change below must remain visually identical to the source image: same pose, same expression, same lighting direction and color temperature, same background, same crop, same camera angle.
SOURCE IMAGE
a founder headshot, seated at a desk, warm window light from the left, plain grey wall behind them
THE ONE CHANGE
change the color of their shirt from pale blue to charcoal grey. Make only this change. Do not use it as license to also adjust skin tone, sharpen the image, recompose the frame, or "improve" anything nearby that wasn't mentioned.
WHAT MUST STAY EXACTLY AS IT IS
their face, expression, hand position, the desk and background, the exact same warm lighting direction and shadow falloff on their face. Treat these as hard constraints, not suggestions — if the requested change would be easier to render by altering one of these, find a way to make the change without touching them instead of trading one for the other.
MATCHING THE EXISTING SCENE
Whatever you add or alter for the one change above must match the existing image's lighting direction, color temperature, and shadow behavior exactly — if the original photo has a warm key light from the upper left, the edited element needs a highlight and shadow consistent with that same light, not generic even lighting. If the change involves a new object or surface at a different depth than what's around it, respect the existing photo's perspective and depth of field rather than rendering the new element in flat, unrelated focus.
WHERE THIS IS GOING
replacing the current LinkedIn profile photo — needs to look like the same photo shoot, not a different day — keep this in mind for how clean and seamless the edit needs to be; a version headed for a quick internal draft has more tolerance for a visible seam than one headed for a print asset or a paid ad.
ITERATION
expect one or two rounds of small correction if the shirt's new color picks up an odd color cast from the background. If the first pass gets the change right but drifts on anything in the "must stay exactly as it is" list, say so specifically in your next message rather than re-describing the whole image from scratch — a targeted correction ("the shadow under the new object is falling the wrong direction, everything else is right") gets a better second pass than a full re-prompt.
OUTPUT
One edited image. If any part of the requested change genuinely cannot be made without visibly affecting something on the preservation list, say so explicitly instead of silently making the trade-off and delivering it as if nothing changed.Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Nano Banana's edit mode operates conversationally on an uploaded image inside one session rather than regenerating from a text prompt alone, and its documented behavior under a vague edit instruction is to sometimes treat the request as license to refine the whole frame — sharpen details, shift color balance, subtly recompose — rather than touch only the named element, which is precisely the failure this prompt heads off by stating up front that this is a single targeted edit and listing, by name, everything that must stay identical. Second, the explicit rule to find another way to make the change rather than quietly trading off something on the preservation list matters because a model under a hard instruction to change X will sometimes resolve an apparent conflict (the new shirt color doesn't sit well under the existing warm light) by adjusting the lighting itself instead of rendering the shirt correctly under the light that's already there — naming that trade-off as forbidden forces the harder, correct solution instead of the easier wrong one. Third, the instruction to match the new element's lighting direction, color temperature, and depth of field to the existing photo is what separates a genuinely seamless single-element edit from an obviously composited one: an added or altered object rendered with flat, generic lighting reads immediately as pasted-in against a photo that has a clear directional key light, and stating the exact light direction the model needs to respect turns an implicit continuity requirement most edit prompts never mention into an explicit, checkable instruction. Finally, structuring iteration as a targeted correction rather than a full re-prompt matters because re-describing the whole scene from scratch on a second attempt effectively restarts the generation with slightly different wording each time, which risks losing whatever the first pass got right — a correction note that names only what drifted keeps the parts that were already correct anchored while the model focuses its next attempt on the one thing that needs fixing.
Verified against
Nano Banana / Gemini 3.1 Flash Image Gemini 3.1 Flash Image · 2026-07-21
Changelog
- 2026-07-21 — Initial publish, verified against Nano Banana (Gemini 3.1 Flash Image) on a single-element clothing-color edit.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
