A repair bench glows at the far end of an otherwise dark workshop. A small delivery robot waits beside it, carrying a parcel that looks too large for its narrow frame. That is the still concept I would develop for a fictional animation project: one image that establishes the room, the character’s scale and the feeling of arriving after everyone else has gone home.
The image-making task belongs before detailed scene construction. I want a visual reference that helps an animator discuss lighting and staging, rather than a finished sequence. In this proposed workflow, nano banana 3 helps turn the workshop brief into a still image and revise its composition. It generates and edits images; animation, rigging and scene assembly would remain separate work in the appropriate production tools.
The robot is an original fictional design for this exercise, with a rounded body, two simple wheels and a shallow parcel tray. I would sketch its silhouette first and keep that sketch available as a reference. Without a clear shape, the workshop’s shelves and tools could become more memorable than the character whose arrival gives the scene a purpose.
Find the relationship between the door and the bench
I would begin with a viewpoint just inside the entrance. The robot sits in the lower third of the image, facing a workbench across the room. Between them is a strip of relatively empty floor. That gap matters: it gives the character somewhere to go and lets the viewer understand the room without tracing a path through clutter.
A first prompt for nano banana 3 might read: “Still concept illustration of a small fictional delivery robot entering a repair workshop at night, view from just inside the doorway, rounded wheeled robot carrying one large parcel, warm task lamp over a workbench at the back, cool dim light near the entrance, clear floor between robot and bench, worn painted cabinets, restrained detail, no readable text, no logos.” The prompt defines a visual relationship rather than trying to specify an entire story.
The parcel should be slightly awkward, but it should not completely hide the robot. If it blocks the body and wheels, the character becomes a cardboard box with an unexplained shadow. My preferred arrangement would show one rounded side of the robot and the edge of its tray. That small overlap can communicate the weight of the delivery without needing a facial expression.
Make the light explain the room
The workbench lamp is the main visual destination. I would keep its warm pool relatively narrow so that the surrounding cabinets remain quiet. A cool entrance light could outline the robot’s upper edge, separating it from the darker floor. Those two light areas should help the viewer move from the arriving character to the place where the parcel belongs.
For a second lighting study in nano banana 3, I might reduce the entrance light and allow the robot to appear mostly as a silhouette. That version could feel more mysterious, but it may also make the parcel difficult to read. Comparing the images at a small size would show whether the mood is coming at the expense of basic staging.
I would avoid adding colourful neon signs merely because the scene happens at night. They could make the workshop look like a different kind of location and introduce additional focal points. The lamp, the doorway and a faint reflection on the floor should be enough. Each new source of light needs a reason within the room, even in a stylised concept.
Choose details that belong to this delivery
The bench might hold an open toolbox and a single empty space where the parcel could be set down. That empty patch connects the room to the character’s task. A wall covered with random gadgets could look impressive while contributing little. I would rather include a few objects whose scale and placement support the imagined action.
The cabinets would have worn paint, but their damage should not overwhelm the image. A couple of scuffed edges can suggest use. Heavy rust, broken doors and piles of debris would imply abandonment, which conflicts with the lit workbench. The concept needs to decide whether this is a functioning repair shop after hours or a forgotten building; mixing both can muddy the story.
Reference-guided editing in nano banana 3 could help explore those surface choices after the broad layout works. I would ask for modest wear on the cabinet fronts while retaining their positions and colour. Then I would inspect the entire room again. An editing request can guide a change, but it does not guarantee that every surrounding object remains untouched.
Correct the first confusing silhouette
Suppose the first image merges the robot’s wheel with the bottom corner of a cabinet. My next revision would create a visible gap between them, perhaps by shifting the cabinet slightly deeper into the room. I would keep the viewpoint and main light unchanged so the effect of that correction is easy to assess.
If the parcel appears to float above the tray, I would request a clearer contact between the two shapes and a small supporting shadow. That is more useful than asking for the image to be “more realistic” without naming the problem. The purpose is not photographic perfection; it is to make the character’s relationship with the object immediately understandable.
Another pass through nano banana 3 might simplify the shelves behind the robot. Small bright objects there could break up its outline, especially at thumbnail size. Removing a few items or darkening that patch may solve the issue more effectively than making the robot brighter. I would prefer a composition that works through placement before relying on an exaggerated glow.
Turn the illustration into a usable reference
Once a direction is chosen, I would prepare a simple reference sheet in a layout application. It would contain the selected image, the original robot sketch and a few short notes about the warm bench light, cool doorway edge and clear floor space. The notes should explain the important decisions without pretending the picture contains measured dimensions or a complete production plan.
The concept is not a source of ready-made geometry. An animator or scene artist would still need to build or select appropriate assets, determine dimensions and establish workable lighting in their own environment. I would treat the image as a conversation about appearance, checking which ideas survive when translated into an actual scene rather than assuming a direct conversion.
For the final handover, I would label the nano banana 3 image as an AI-generated concept and keep the prompt and revision notes beside it. If another artist changes the robot design, the concept can be revised around the new reference. Its value lies in making the proposed mood and staging visible early, while those decisions are still easy to discuss.
The scene does not need to reveal what is inside the parcel. A lit bench, a waiting robot and a clear strip of floor already establish a small unanswered question. My aim would be to preserve that simplicity in the still image, giving the next stage of creative work a specific place to begin.