A practical prompt reference for photographers and content creators using Nano Banana 2 (Gemini 3.1 Flash Image) to change camera angles without reshooting.
1. What This Guide Covers
Nano Banana 2 (technically Gemini 3.1 Flash Image) is Google's latest image generation and editing model. It combines the advanced spatial reasoning of Nano Banana Pro with the speed of the Gemini Flash architecture, meaning it can understand 3D relationships within a 2D image and re-render scenes from new viewpoints.
This guide focuses on one specific capability: taking an existing photograph and shifting the camera perspective – moving the virtual camera left, right, higher, lower, or to a completely different position – while preserving the subject's identity, clothing, lighting, and environment.
This is especially useful for photographers, video creators, and marketing professionals who have a single shot and need to simulate alternative camera positions without reshooting.
2. How Nano Banana 2 Understands Perspective
Unlike traditional image editors that warp pixels, Nano Banana 2 reasons about the spatial structure of a scene before generating the output. It analyses depth, lighting direction, object geometry, and spatial relationships to reconstruct what the scene would look like from a different vantage point.
Key capabilities that make perspective shifting possible:
- Spatial reasoning: The model understands 3D relationships – which objects are in front, behind, above, or below – and adjusts parallax, foreshortening, and occlusion accordingly.
- Subject preservation: It maintains identity consistency for up to 5 characters, keeping facial features, clothing, and accessories stable across angle changes.
- Lighting inference: When you move the camera, the model recalculates how light falls on surfaces from the new vantage point, adjusting highlights, shadows, and reflections.
- Configurable thinking: You can set the thinking level to “High” for complex perspective changes, giving the model more reasoning time before rendering.
3. The Perspective Shift Prompt Formula
For consistent results when shifting perspective on an existing image, structure your prompt using this four-part framework:
- ANCHOR – What must stay the same (subject, outfit, setting, lighting style)
- CAMERA MOVE – Where the camera is now moving to (direction, height, distance)
- SUBJECT RESPONSE – How the subject should adjust (gaze direction, body orientation)
- TECHNICAL CONSTRAINTS – Lens, depth of field, and any preservation instructions
The Formula in Practice
Here is the general template you can adapt for any scenario:
Template Prompt:
Keep everything in this image exactly as it is – same [subject description], same [outfit/props], same [environment], same [lighting quality]. Move the camera to [new position relative to current]. The subject is now [looking at / turned toward] the new camera position. Shot with [lens/focal length], [depth of field]. Maintain consistent proportions, shadows, and colour grading.
Why this works: Anchoring first prevents the model from reimagining the entire scene. Camera terminology early in the prompt ensures the spatial instruction takes priority.
⚠ Pro Tip: Place your angle and lens instructions in the first third of the prompt. Nano Banana 2 prioritises spatial composition instructions that appear early, before style and texture descriptors.
4. Worked Example: News Presenter Perspective Shift
Scenario: You have a photograph of a news presenter sitting behind a news desk. The current shot is straight-on (front-facing). You need to move the camera to the right side of the set so the presenter appears in a three-quarter view, now looking directly at the new camera position.
Step 1: Describe What Stays the Same
Start by locking the elements that must not change:
Anchor Prompt:
Keep the exact same news presenter, same face, same hairstyle, same suit and tie, same news desk, same studio background with monitors, same studio lighting setup with key light from the left and fill light from the right.
Why this works: Being specific about lighting positions is crucial – when the camera moves, the model needs to know where the lights are so it can adjust highlights and shadows correctly.
Step 2: Define the Camera Movement
Now describe where the camera is moving to. Be explicit about direction and approximate angle:
Camera Move Prompt:
Move the camera approximately 30–45 degrees to the right of the current straight-on position, keeping the camera at the same height (eye level with the seated presenter). The camera distance from the subject remains the same as the original shot.
Why this works: Specifying the degree of rotation gives the model a measurable target. Keeping height and distance constant prevents unwanted scale or perspective distortion changes.
Step 3: Adjust the Subject
The presenter needs to look at the new camera position rather than continuing to face the old one:
Subject Response Prompt:
The presenter has turned their head and upper body slightly to face the new camera position directly. They are making direct eye contact with the camera. Their posture remains professional and natural – seated upright behind the desk.
Why this works: Without this instruction, the model may keep the subject facing the original direction, resulting in an awkward profile shot rather than the engaged look you need.
Step 4: Add Technical Constraints
Technical Prompt:
Shot with a 50mm lens at f/2.8, shallow depth of field with the presenter sharp and the studio background slightly soft. Maintain the same colour grading, white balance, and broadcast-quality lighting. Do not change any studio set elements, graphics, or desk props.
Why this works: Lens and aperture instructions help the model maintain the photographic feel of the original rather than drifting toward an illustration style.
Complete Combined Prompt
Here is the full prompt combining all four parts into a single, cohesive instruction:
Full Prompt – News Presenter Right-Angle Shift:
Keep the exact same news presenter – same face, hairstyle, suit and tie – sitting behind the same news desk in the same studio with monitors in the background. Same studio lighting with key light from camera-left and fill from camera-right. Move the camera approximately 35 degrees to the right of the current straight-on position, maintaining eye-level height and the same distance from the subject. The presenter has turned their head and upper body to face the new camera position directly, making natural eye contact. Seated posture remains professional and upright. Shot with 50mm lens, f/2.8 shallow depth of field, presenter sharp, background slightly soft. Maintain consistent colour grading, white balance, and broadcast lighting quality throughout. Do not alter any studio set elements, desk props, or background graphics.
Why this works: This prompt runs approximately 120 words – well within Nano Banana 2's comprehension sweet spot. The model follows complex, multi-layered instructions more reliably than many competing models.
5. More Perspective Shift Prompt Examples
The following examples cover common scenarios where you'd want to shift the camera on an existing image. Each follows the same four-part structure.
5a. Product on a Table – Shift to Three-Quarter Overhead
Scenario: A flat-lay product shot taken directly from above needs to be shifted to a three-quarter overhead angle to add depth.
Keep the exact same product, same packaging, same table surface and props in their current positions. Same soft diffused overhead lighting. Shift the camera from the current top-down position to a three-quarter overhead angle (approximately 45 degrees from vertical), viewing the product from the front-right. The product labels and branding text should still be clearly readable. Shot with 50mm lens, f/4 medium depth of field keeping the product and immediate surroundings sharp. Maintain the same colour palette, shadow softness, and editorial styling. Do not add or remove any objects.
Why this works: Mentioning “labels should still be readable” gives the model a legibility constraint that prevents it from rotating the product to an angle where text becomes obscured.
5b. Architectural Interior – Shift Camera Left Along the Wall
Scenario: An interior photograph shot from a doorway looking into a room needs the camera moved to the left to reveal more of the right-hand wall.
Keep the exact same room – same furniture, same wall colours, same flooring, same window light coming from the right. Move the camera approximately 2 metres to the left of the current position, keeping the same height (eye level, standing). The camera should now be angled slightly to the right to keep the centre of the room in frame, revealing more of the right-hand wall and the window. Shot with 24mm wide-angle lens, f/8 deep focus keeping the entire room sharp. Maintain the same natural daylight colour temperature and shadow direction. Do not change any furniture positions or decorative elements.
Why this works: Describing the move in real-world units (metres) and specifying the compensating angle adjustment helps the model produce a natural-feeling shift rather than a mechanical rotation.
5c. Portrait – Shift from Front-Facing to Three-Quarter View
Scenario: A headshot taken straight-on needs to be shifted to a classic three-quarter portrait angle.
Keep the exact same person – same face, same expression, same hairstyle, same clothing, same background. Same soft studio lighting with key light from camera-left. Move the camera approximately 30 degrees to the right so the face is seen in a classic three-quarter portrait view. The subject's eyes should follow the camera, maintaining direct eye contact with the new position. Shot with 85mm portrait lens, f/1.8 very shallow depth of field, face tack-sharp, background fully blurred. Keep the same skin tones, colour grade, and overall mood. Do not change the subject's clothing, accessories, or expression.
Why this works: The 85mm lens instruction creates the flattering compression typical of portrait photography, and the shallow f/1.8 aperture reinforces the portrait aesthetic.
5d. Street Scene – Shift from Eye Level to Low Angle
Scenario: A street-level photograph of a building needs the camera dropped to a low angle looking upward to create a more dramatic, imposing feel.
Keep the exact same building, same street, same pedestrians in their current positions, same overcast daylight. Drop the camera to a low angle, approximately knee height, tilted upward toward the top of the building. The building should now appear to tower above the viewer with converging vertical lines. Shot with 24mm wide-angle lens, f/5.6, deep focus keeping both the street foreground and building sharp. Maintain the same overcast lighting, muted colour palette, and urban atmosphere. Do not add or remove people or vehicles.
Why this works: Mentioning “converging vertical lines” tells the model to apply realistic perspective distortion rather than keeping the verticals parallel as in a corrected architectural shot.
5e. Group Shot – Shift Camera to the Left, Subjects Track
Scenario: A group photo taken from centre needs the camera moved to the left, with all subjects turning to face the new camera position.
Keep the exact same group of people – same faces, same clothing, same standing positions, same outdoor park background with trees. Same natural afternoon sunlight from the right. Move the camera approximately 25 degrees to the left. All subjects should naturally turn their heads and bodies slightly to the right to face the new camera position, maintaining relaxed, natural smiles. The group composition and spacing should remain the same. Shot with 35mm lens, f/4, medium depth of field with the group sharp and the trees softly blurred. Maintain the same warm afternoon colour temperature and shadow direction. Do not change any clothing, accessories, or background elements.
Why this works: Specifying that subjects should turn “slightly to the right” (opposite to the camera's leftward move) ensures they track the camera naturally rather than all turning in the wrong direction.
5f. Food Photography – Shift from Overhead to 45-Degree Angle
Scenario: A flat-lay food photograph taken from directly above needs to be reframed at a 45-degree dining angle to make the dish more appetising.
Keep the exact same dish, same plate, same table setting, same garnishes in their current positions. Same soft window light from camera-left. Shift the camera from the current top-down position to a 45-degree angle, viewing the plate from the front. The dish should now show depth and dimension – you can see the sides of the plate and the height of the food. Shot with 85mm lens, f/2.2 shallow depth of field with the front of the dish sharp and the background cutlery softly blurred. Maintain the same warm, appetising colour grading and soft shadow quality. Do not rearrange any food items or table elements.
Why this works: The instruction to “show depth and dimension” and “see the sides of the plate” reinforces the spatial change the model needs to achieve, beyond just specifying the angle number.
6. Advanced Techniques & Troubleshooting
6a. Using Thinking Levels
Nano Banana 2 supports configurable thinking levels. For complex perspective shifts – especially those involving multiple subjects, intricate environments, or large angular changes – set the thinking level to “High” or “Dynamic”. This gives the model additional reasoning time to plan the spatial reconstruction before rendering, significantly improving accuracy on difficult shots.
6b. Iterative Refinement
Nano Banana 2 supports conversational editing. If the first result isn't quite right, you can follow up with refinement prompts without re-uploading the image:
“Shift the camera another 10 degrees further to the right.”
“The presenter's gaze is slightly off – adjust so they are looking directly into the lens.”
“The shadows on the desk are too harsh for the new angle – soften them to match the original lighting feel.”
“Pull the camera back slightly to include more of the desk in the frame.”
Why this works: Conversational follow-ups are processed in context, so the model remembers what it did on the previous generation and adjusts incrementally.
6c. Common Problems and Fixes
| Problem | Solution |
|---|---|
| Subject's face changes identity | Add more specific anchoring: “same face, same exact facial features, same skin tone” and use the reference image feature. |
| Lighting doesn't match new angle | Explicitly state the light source positions relative to the scene (not the camera): “key light is positioned stage-left, fill light stage-right.” |
| Proportions feel distorted | Specify a standard focal length (50mm or 85mm) and avoid extreme angles (beyond 60 degrees) in a single step. |
| Background elements shift incorrectly | Name specific background elements: “the monitor showing the station logo stays in the same position on the back wall.” |
| Subject still facing original camera | Add explicit gaze direction: “the subject's eyes and head have turned to look directly into the new camera position.” |
| Too much change overall | Break large shifts into two steps: first move 20 degrees, then use the result as input and move another 20 degrees. |
6d. Optimal Image Quality for Input
The quality of your perspective shift depends heavily on the input image. For best results:
- Use the highest resolution available – Nano Banana 2 supports up to 5MB uploads in JPEG, PNG, or WebP.
- Sharp, well-exposed originals produce better spatial reconstruction than soft or underexposed images.
- Images with clear depth cues (foreground/background separation, visible surfaces receding into distance) give the model more spatial information to work with.
- Avoid heavily cropped images where important context has been removed – the model needs environmental clues to reconstruct the scene from a new angle.
7. Quick-Reference Prompt Vocabulary
These are the most effective terms for communicating camera moves to Nano Banana 2. Use them at the beginning of your prompt for the strongest effect:
| Camera Term | What It Communicates |
|---|---|
| Move camera to the right / left | Horizontal orbit around the subject |
| Drop camera to low angle | Camera below subject, looking up |
| Raise camera to high angle | Camera above subject, looking down |
| Bird's-eye view / top-down | Directly overhead, looking straight down |
| Worm's-eye view | Camera on the ground, looking straight up |
| Three-quarter view | Approximately 30–45 degrees off the front-facing axis |
| Dutch angle / tilted horizon | Camera rotated on its roll axis for dynamic tension |
| Pull back / dolly out | Increase distance from subject |
| Push in / dolly in | Decrease distance, tighter framing |
| Crane up / crane down | Vertical camera movement while keeping subject in frame |
| Over-the-shoulder | Camera positioned behind one subject looking at another |
| 50mm / 85mm / 24mm lens | Controls perspective compression and field of view |
8. Summary
The key to reliable perspective shifts in Nano Banana 2 is structured prompting: anchor what stays, describe the camera move, direct the subject's response, and constrain the technical look. The model's spatial reasoning is remarkably capable, but it performs best when given clear, specific, cinematography-informed instructions.
Start with modest angle changes (15–30 degrees) and work up to larger shifts as you develop a feel for how the model interprets your instructions. Use conversational refinement to dial in the exact result, and remember that breaking a large move into two smaller steps often produces more consistent results than attempting a dramatic shift in one pass.