Verified against Nano Banana / Gemini 3.1 Flash Image · 2026-07-22
Keep one character consistent across a set of different scenes
Locks a defined character's face, proportions, and defining features across a run of separately-described scenes in the same Nano Banana conversation, instead of letting each new generation quietly redesign the character.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are generating a set of illustrations or photos that must all depict the exact same character, a brand mascot: a small orange fox wearing round wire glasses and a forest-green scarf. Treat every scene below as a costume and setting change for one fixed individual — never as an opportunity to redesign their face, build, or defining features. CHARACTER — LOCKED FEATURES These features must be identical in every single scene, with no drift between them: the round wire glasses, the exact scarf color and knot style, the white chest-fur patch, and the slightly oversized left ear. If a scene's pose or angle makes one of these features partly obscured or hard to see, keep it consistent with how it would look from that angle on this specific character — don't substitute a generic version because the exact feature is harder to render at that angle. SCENES TO GENERATE 1. Sitting at a laptop in a bright home office. 2. Walking through a farmer's market carrying a paper bag. 3. Waving from behind a podium at a small conference.. Generate these as a numbered set, referring back to this same character description for every one — do not let later scenes in the set drift toward a "cleaner" or subtly different version of the character than the first one you generated. POSE AND EXPRESSION friendly and approachable in every scene, but the specific pose and hand position should suit each scene's action naturally for each scene as specified — expression and pose should vary scene to scene where the brief calls for it, but the underlying face structure, proportions, and defining features must not change just because the expression did. STYLE CONSISTENCY flat, warm-toned vector-illustration style with soft cel-shaded shadows, consistent line weight throughout. Every scene in this set needs to look like it came from the same rendering pass — same line quality or photographic treatment, same general lighting philosophy — even though the settings differ, so this reads as one coherent character across a story or campaign rather than several unrelated images that happen to share a name. FRAMING Render each scene at 4:5, matching an Instagram carousel, framed appropriately for what's happening in that specific scene, while keeping the character recognizably the same size and prominence relative to frame unless a scene specifically calls for a wider or tighter shot. CHECK BEFORE DELIVERING Before finalizing each scene, compare the character against the locked features list above and against the first scene you generated in this set — if anything drifted (a different eye shape, a hairstyle detail that changed, proportions that shifted), regenerate that one scene rather than delivering a set with a visible inconsistency in it. OUTPUT The full numbered set of scenes, each as its own image, plus a one-line note on any feature you found genuinely difficult to keep perfectly consistent across the set and why.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Nano Banana retains a running visual context within a single conversation thread rather than starting from a blank slate on every message, which is the documented mechanism that makes character consistency across a set of prompts even possible in the first place — but that same session memory degrades gradually across a long run of separately-worded scene descriptions unless each one is explicitly anchored back to the same locked feature list, which is why this prompt repeats the character's defining features as a named, comparable checklist rather than trusting the model to remember an earlier description verbatim ten scenes later. Second, the instruction not to substitute a "cleaner" generic version when a feature is partly obscured by pose or angle targets a specific and common drift pattern: a model asked to render the same round-glasses, orange fox character from behind, where the glasses are barely visible, will sometimes default to a simpler, more generic fox rendering for that one frame rather than working out how this specific character's proportions would actually look from an unusual angle, and the result is a set where one scene subtly doesn't match the others even though nothing was technically changed on purpose. Third, requiring an explicit self-check against both the locked-feature list and the first generated scene — rather than trusting a single generation pass to get it right — matters because consistency errors in multi-image sets compound directionally: a slight drift in scene two, uncorrected, becomes the new implicit reference for scene three, and by scene five the character has walked measurably away from where it started without any single step looking obviously wrong in isolation, which is exactly the kind of gradual failure that only becomes visible when someone lines up the whole set side by side after delivery instead of catching it mid-generation.
Verified against
Nano Banana / Gemini 3.1 Flash Image Gemini 3.1 Flash Image · 2026-07-22
Changelog
- 2026-07-22 — Initial publish, verified against Nano Banana (Gemini 3.1 Flash Image) on a three-scene brand-mascot set.
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
