Fixing Inconsistent Lighting on Multilingual Labels in Nano Banana 2

When generating or editing product imagery with Nano Banana, users often encounter a specific visual artifact: the lighting on newly added multilingual text does not match the ambient environment of the curved packaging surface. This issue typically manifests as text that appears flat, overly bright, or cast with shadows that contradict the direction of the primary light source on the container. For example, if a bottle is lit from the top-left, the text might appear illuminated from the front or have no shadow at all, breaking the realism of the image.

This symptom is particularly common when working with complex geometries like cylindrical jars or angled boxes where the curvature affects how light wraps around the object. The inconsistency arises because the AI model must simultaneously interpret the existing surface lighting and apply new typography that adheres to those same physical laws. When these two elements are not perfectly synchronized in the prompt instructions, the result is a label that looks pasted on rather than integrated into the scene.

Distinguishing Symptoms from Known Model Behaviors

To effectively troubleshoot this issue, it is crucial to separate the observed symptom from the known capabilities and limitations of the underlying technology. The symptom is strictly visual: a mismatch between the illumination of the background packaging and the foreground text.

It is important to note that Nano Banana refers to the AI image generation and editing tool, not a skincare brand or physical product. While the tool supports text-to-image and image-to-image workflows, its prompt instructions describe desired outcomes but do not guarantee identity, label, object, or typography preservation. This means that even with detailed prompts, the model may struggle to maintain perfect consistency across different languages or complex curves without specific guidance.

Furthermore, the platform hosts distinct Google image models. Nano Banana 2 corresponds to Gemini 3.1 Flash Image, while Nano Banana Pro uses Gemini 3 Pro Image. These are distinct models with varying strengths. Users should be aware that Nano Banana 2 Lite, which corresponds to Gemini 3.1 Flash Lite Image, is focused on speed and cost. It is explicitly not optimized for multiple reference inputs or multi-turn sequential editing. Therefore, attempting to fix complex lighting issues on multilingual labels using the Lite version may yield inconsistent results compared to the standard Nano Banana 2 or Pro versions. Do not assume features available in the Pro model are present in the Lite version without verifying the specific workflow requirements.

Diagnosing the Root Cause of Lighting Mismatches

The root cause of inconsistent lighting usually lies in the ambiguity of the prompt regarding the light source's position, intensity, and color temperature relative to the text. When adding multilingual text, the model may default to a neutral, flat lighting setup rather than calculating the specular highlights and soft shadows required by the curved surface.

Another factor is the complexity of the language itself. Different scripts (e.g., Latin, Cyrillic, or CJK characters) have different stroke widths and spacing. If the prompt does not explicitly account for how these variations interact with the surface curvature, the model might render the text with uniform depth, ignoring the perspective distortion caused by the curve. This leads to a scenario where the text looks correct in isolation but fails to blend with the surrounding geometry.

Additionally, the lack of explicit negative constraints can contribute to the problem. Without telling the model what not to do (such as avoiding flat shading or harsh contrast), the generator might prioritize legibility over environmental integration. The model needs clear directives to treat the text as an object subject to the same physics as the bottle or box.

Adjusting Prompts to Match Ambient Light Sources

Correcting lighting mismatches requires precise adjustments to your prompt parameters. Start by defining the ambient light source with high specificity. Instead of simply saying "lit," specify the angle, such as "soft overhead lighting casting shadows to the bottom right." Explicitly state that the text must follow the curvature of the surface.

You can enhance the prompt by describing the interaction between the light and the material. For instance, add phrases like "specular highlights on the curved glass" or "subsurface scattering on the paper label." This helps the model understand the texture and how light should behave on the specific medium. When dealing with multilingual text, ensure the prompt mentions that all characters, regardless of script, must adhere to the same lighting rules.

If you are using the Nano Banana 2 interface, consider utilizing the prompt library for inspiration. The site offers example prompts that users can copy or take into the generator. Look for examples that involve curved surfaces or complex lighting scenarios. However, remember that these are examples; they do not guarantee identity or typography preservation. You will likely need to adapt them to your specific multilingual context.

For more complex edits involving multiple references or sequential changes, ensure you are using the appropriate model version. As noted, Nano Banana 2 Lite is not optimized for these workflows. If you find the lighting still inconsistent after adjusting the prompt, switching to Nano Banana Pro (Gemini 3 Pro Image) might provide better control over the fine details of the lighting and text integration.

Verifying the Fix and Finalizing Your Image

Once you have adjusted your prompt, generate the image and verify the result against the original lighting conditions. Check if the shadows on the text align with the shadows on the container. Look for specular highlights that match the direction of the light source. If the text appears too flat or the shadows are in the wrong direction, refine the prompt further by adding more descriptive adjectives about the light quality.

It is also helpful to compare the generated text against the curvature of the object. The letters should appear slightly distorted if they are wrapping around a cylinder, just as the rest of the label does. If the text looks perfectly straight on a curved surface, the lighting and perspective are likely incorrect.

Remember that prompt instructions describe desired outcomes but do not guarantee identity or typography preservation. There may be instances where the model struggles to perfectly integrate complex scripts with difficult lighting. In such cases, iterative refinement is key. Small adjustments to the lighting description often yield significant improvements in the final output.

By understanding the distinction between the tool's capabilities and the specific visual challenges of multilingual labels, you can systematically address lighting inconsistencies. Whether you are working with simple English text or complex international scripts, precise prompt engineering is the most effective way to achieve realistic results. For more information on the tools and models available, visit Try Nano Banana.

Always refer to the official documentation for the latest updates on model capabilities, as features and performance can evolve. By following these troubleshooting steps, you can ensure your product imagery maintains professional quality and visual coherence across all languages and surfaces.