Mastering Perspective Correction with Nano Banana 2 Image Prompts

When photographing architecture or urban environments, it is common to encounter perspective distortion. Vertical lines may lean inward, and horizontal planes might appear tilted due to the camera angle. While traditional editing software requires manual manipulation of layers and masks, Nano Banana offers a streamlined approach through its image-to-image workflow. By leveraging specific prompt structures, you can guide the AI to re-align geometric elements without altering the fundamental content of the scene.

This tutorial focuses on the Nano Banana 2 model, identified by Google as Gemini 3.1 Flash Image. This specific model supports text-to-image and image-to-image workflows, making it suitable for tasks where you need to maintain the visual identity of an input image while correcting its geometry. It is important to note that prompt instructions describe desired outcomes but do not guarantee the preservation of specific labels, typography, or exact object identities. Therefore, the goal here is structural alignment rather than perfect replication of every pixel detail.

Structuring Prompts for Geometric Alignment

The core of successful perspective correction lies in how you articulate the desired geometric changes within your prompt. Generic requests like "fix the photo" often yield inconsistent results because they lack directional guidance. Instead, you must explicitly define the relationship between vanishing points and the horizon line.

To achieve this, your prompt should prioritize terms related to architectural stability and linear alignment. Focus on describing the ideal state of the building or object rather than the current error. For instance, instead of saying "make the building straight," use phrasing that emphasizes verticality and parallelism. You might include instructions such as "align vertical lines to be perfectly perpendicular to the ground plane" or "correct the perspective so that all vertical edges converge at a single vanishing point."

These instructions act as a directive for the model to reinterpret the spatial relationships within the image. The AI analyzes the input image's composition and applies these geometric constraints during generation. When using the Nano Banana tool, ensure your prompt clearly distinguishes between the content you want to keep (the textures, colors, and objects) and the geometry you wish to change (the angles and alignment). This separation helps the model understand that the distortion is a technical flaw to be fixed, not a stylistic choice to be preserved.

Step-by-Step Workflow for Image-to-Image Correction

Executing a perspective fix requires a systematic approach to ensure the best possible output. Follow these numbered steps to utilize the Nano Banana 2 interface effectively:

1. Upload Your Source Image: Begin by selecting the image containing the perspective distortion. Ensure the file format is supported by the platform and that the subject matter is clearly visible.
2. Select the Model: Confirm that you are using the Nano Banana 2 engine. While other versions exist, such as Nano Banana Pro or Nano Banana 2 Lite, this specific task benefits from the capabilities of the standard Nano Banana 2 model. Note that Nano Banana 2 Lite is focused on speed and cost and is not optimized for complex multi-turn sequential editing or multiple reference inputs, so it may not provide the same level of control for detailed geometric corrections.
3. Compose the Prompt: Enter your structured prompt into the text field. Use the guidelines mentioned above to emphasize vanishing points and vertical alignment. Remember that examples provided in the prompt library are untested templates; you should adapt them to your specific image needs.
4. Generate and Review: Submit the request and observe the generated result. Compare the new image against the original to see if the lines have been realigned correctly.
5. Iterate if Necessary: If the first result does not fully correct the perspective, refine your prompt. Add more specific details about the orientation of the walls or the position of the horizon line. You may also try adjusting the strength of the image-to-image influence if the tool provides such a slider.

For those looking to experiment immediately, you can Try Nano Banana to access the generator and apply these techniques to your own photos.

Evaluating Results and Troubleshooting Common Issues

Judging the success of a perspective correction involves checking both the geometry and the integrity of the scene. A successful output will show vertical lines that appear truly vertical and horizontal lines that run parallel to the bottom edge of the frame. However, you must also verify that the content has not been inadvertently altered. Sometimes, aggressive perspective correction can cause warping in textures or the introduction of artifacts where the AI struggles to fill in the gaps created by the geometric shift.

If the lines remain skewed, your prompt may lack sufficient emphasis on the vanishing point. Try adding explicit references to the "horizon line" or "ground plane" to give the model a stronger anchor. Conversely, if the image looks overly smoothed or loses texture detail, your prompt might be too dominant over the source image. In this case, consider softening the language or reducing the weight of the geometric instructions.

It is crucial to remember that Nano Banana does not guarantee identity or label preservation. Text within the image, such as street signs or storefront names, may be altered or removed during the process. This is a known limitation of the prompt system, which prioritizes the overall visual outcome over specific textual fidelity. If you require precise text retention, additional post-processing or different tools may be necessary. By understanding these limitations and refining your prompt structure, you can consistently achieve professional-grade perspective corrections using the Nano Banana 2 image-to-image workflow.