Fixing Inconsistent Button Placement on Victorian Waistcoats in Nano Banana 2

When generating historical fashion imagery, precision is paramount. A Victorian waistcoat relies heavily on symmetry and structured tailoring to convey authenticity. However, users of the AI image generation tool Nano Banana often encounter a specific visual artifact: buttons that appear at varying heights or tilted at odd angles across a single garment. This inconsistency breaks the illusion of a tailored piece of clothing, making the output look flawed rather than historically accurate. This guide addresses how to diagnose this symptom, separate plausible causes from known facts, and apply specific prompt engineering techniques to achieve strict vertical alignment.

Understanding the Symptom and Known Facts

The primary symptom involves the misalignment of small details within a complex texture. Instead of a straight, vertical column of buttons running down the center of the waistcoat, the generated image displays buttons that drift left or right, sit at different vertical intervals, or rotate slightly off-axis. This issue is particularly noticeable when the garment occupies a significant portion of the frame or when the lighting creates strong shadows that confuse the model's depth perception.

It is crucial to distinguish between user expectations and the current capabilities of the underlying technology. Google documents Nano Banana 2 as Gemini 3.1 Flash Image (gemini-3.1-flash-image). While powerful, prompt instructions describe desired outcomes but do not guarantee identity, label, object, or typography preservation. This means that while you can ask for "perfectly aligned buttons," the model does not possess a built-in ruler or geometric constraint engine that enforces pixel-perfect placement without iterative refinement. Furthermore, Nano Banana refers to the AI image generation/editing tool; it is not a skincare brand, bottle, jar, or physical subject. The tool operates on text-to-image and image-to-image workflows, but it does not inherently understand the structural rigidity of fabric unless explicitly guided through detailed prompting.

Separating Plausible Causes from Model Limitations

To fix the issue, one must first identify whether the problem stems from the prompt ambiguity or the model's inherent limitations regarding spatial reasoning. A common plausible cause is the lack of explicit directional constraints in the positive prompt. If a user simply requests a "Victorian waistcoat with buttons," the model may prioritize the overall aesthetic of the fabric over the geometric arrangement of the fasteners. The model might interpret "buttons" as a general texture pattern rather than discrete objects requiring linear alignment.

Another factor to consider is the version of the model being used. Google describes Nano Banana 2 Lite as focused on speed and cost. It is not optimized for multiple reference inputs or multi-turn sequential editing. If a user attempts to force high-fidelity alignment using Nano Banana 2 Lite, they may face more frequent inconsistencies because the model sacrifices some detail fidelity for processing speed. Additionally, the website has a Nano Banana Pro page at /nanobananapro, which utilizes Gemini 3 Pro Image (gemini-3-pro-image). Users should be aware that Google model names and capabilities must not be presented as proof of identical features on this website, so feature availability varies by plan.

It is also important to note that the prompt library offers example prompts that users can copy or take into the generator. These examples are untested in real-time scenarios and serve only as starting points. They do not guarantee that a specific prompt will result in perfect button alignment every time. Therefore, relying solely on pre-written examples without adjusting for specific alignment needs is a likely cause of failure.

Diagnosing and Fixing Alignment Issues

Diagnosing the root cause involves analyzing the generated output against the prompt structure. If the buttons are present but misaligned, the issue is likely a lack of negative constraints or insufficient descriptive density regarding geometry. To fix this, you must adjust negative prompts to enforce strict vertical alignment constraints. Negative prompts act as a filter, telling the model what not to generate. By explicitly stating that buttons should not be crooked, uneven, or scattered, you guide the generative process toward symmetry.

For instance, instead of just asking for a waistcoat, refine the prompt to include phrases like "symmetrical button row," "straight vertical alignment," and "even spacing." Simultaneously, use negative prompts such as "crooked buttons," "tilted fasteners," "asymmetrical layout," and "uneven button height." This dual approach reinforces the desired geometry. If the initial results still show minor deviations, try increasing the weight of the alignment descriptors in your prompt or switching to a higher-tier model if available, as the base model may struggle with fine-grained spatial logic.

Remember that these adjustments require iteration. There is no single command that guarantees a perfect result immediately. You may need to run multiple generations, tweaking the phrasing of the alignment constraints each time. For users looking to experiment with these advanced prompt structures, Try Nano Banana provides the interface to test these variations directly.

Verifying the Solution

Once you have applied the refined prompts and negative constraints, verification is the final step. Inspect the generated image closely for the consistency of the button line. Do the buttons form a straight vertical axis? Is the spacing between them uniform? Are they perpendicular to the ground plane of the image? If the buttons align correctly and the waistcoat appears structurally sound, the troubleshooting process is successful. If inconsistencies remain, revisit the negative prompt list to ensure no contradictory terms were included and consider whether the resolution of the image is too low for the model to render small details accurately. By systematically applying these constraints, users can significantly improve the structural integrity of their generated Victorian garments.