Fixing Uneven Lighting in Nano Banana 2 Batch Product Shots

When creating a catalog of product images, consistency is paramount. A common challenge users face with batch generation workflows in Nano Banana 2 is encountering uneven lighting between shots. You might notice that one variation has a shadow falling to the left while another casts it to the right, or that the overall brightness fluctuates wildly across the set. This inconsistency can disrupt the visual flow of an e-commerce store, making products appear as if they were photographed under different conditions rather than generated by a single system.

This issue typically stems from how the AI interprets randomization parameters and how specific lighting cues are weighted within the prompt structure. Understanding the distinction between the tool's capabilities and the variables you control is the first step toward a solution. Nano Banana refers to the AI image generation and editing tool used here; it is not a skincare brand, bottle, jar, or physical subject. The goal is to standardize the output so your product line looks cohesive without manually editing every single file.

Distinguishing Symptoms from Known Facts

Before attempting a fix, it is crucial to separate the observable symptoms from the underlying technical facts provided by the model documentation. The symptom is clear: batch-generated images of the same product exhibit varying shadow directions, intensity levels, and color temperatures. This creates a disjointed look where the lighting environment feels unstable.

However, known facts about the system clarify what is happening under the hood. Google documents Nano Banana 2 as Gemini 3.1 Flash Image (gemini-3.1-flash-image). While this model supports text-to-image and image-to-image workflows, prompt instructions describe desired outcomes but do not guarantee identity, label, object, or typography preservation. This means that even with identical prompts, slight variations in how the model renders light sources can occur due to its probabilistic nature.

It is also important to note that the website hosts a Nano Banana Pro page at /nanobananapro and a Nano Banana Lite page at /nanobananalite. However, the existence of these pages does not automatically establish support for all Google model names and capabilities on this specific site. For instance, Google describes Nano Banana 2 Lite as focused on speed and cost, noting it is not optimized for multiple reference inputs or multi-turn sequential editing. Therefore, relying on Lite versions for complex batch lighting tasks without understanding these limitations could exacerbate inconsistencies. Stick to the standard Nano Banana 2 workflow for best results regarding lighting stability.

Diagnosing the Root Cause of Inconsistency

The primary driver of uneven lighting in batch generation is often the lack of deterministic controls in the prompt or seed values. When you generate multiple images using a generic prompt like "a white sneaker on a gray background," the AI introduces randomness to create variety. Without explicit constraints, the direction of the light source becomes a variable rather than a constant.

Another factor is the ambiguity of lighting descriptors. Terms like "bright" or "soft light" are subjective to the model's training data. One iteration might interpret "bright" as high-key studio lighting, while another applies a softer, diffused glow. Furthermore, if you are using image-to-image workflows, the reference image itself might contain subtle lighting cues that the model amplifies differently in subsequent generations. If the initial reference has a specific shadow angle, the model might try to preserve it but fail to maintain it perfectly across a large batch, leading to drift.

To diagnose this, review your prompt history. Are you using vague terms? Are you changing the seed value randomly between batches? Are you inadvertently mixing models that have different rendering characteristics? Identifying these patterns helps isolate whether the issue lies in the prompt engineering or the parameter selection.

Fixing Lighting Variations with Seeds and Descriptors

Standardizing your lighting requires a two-pronged approach: refining your textual descriptors and controlling the randomization seeds. Start by replacing vague lighting terms with precise, directional language. Instead of saying "nice lighting," specify "single key light from the top-left at 45 degrees, softbox diffusion." By defining the angle, type, and quality of the light source explicitly, you reduce the model's freedom to invent conflicting environments.

Next, address the seed values. In batch generation, if you want consistent lighting, you should generally use the same seed value for all variations unless you specifically need to alter the product pose or texture. Using a fixed seed ensures that the underlying noise pattern remains constant, allowing the model to focus on applying the same lighting logic to different product attributes. If you must vary the seed for creative reasons, ensure your prompt includes a strong anchor for the lighting setup that overrides the random noise.

For users looking to experiment with these settings, you can explore the prompt library which offers example prompts that users can copy or take into the generator. These examples often demonstrate how to structure lighting commands effectively. Remember, prompt instructions describe desired outcomes; they do not guarantee identity, label, object or typography preservation, so always verify the final output against your brand standards.

If you find that the standard workflow still yields inconsistent results, consider testing the Try Nano Banana interface to see how different model configurations handle your specific lighting requests. Note that while Nano Banana 2 is robust, it is distinct from Nano Banana 2 Lite, which is focused on speed and cost and may struggle with maintaining complex lighting consistency across multiple turns.

Verifying Consistency Across Your Catalog

Once you have adjusted your prompts and seed values, verification is the final step. Generate a small test batch of five to ten images using your new standardized settings. Compare them side-by-side to check for shadow direction, brightness levels, and color temperature. Look for any lingering drift where one image appears significantly darker or has a shadow pointing in a different direction.

If the test batch shows uniformity, proceed with generating the full catalog. If inconsistencies persist, revisit your descriptors. Perhaps the lighting was too complex for the model to render consistently in a single pass. Try simplifying the scene description or breaking the task into smaller, more controlled steps. Consistency in AI generation is an iterative process. By carefully managing your inputs and understanding the model's limitations, you can achieve professional-grade lighting uniformity for your e-commerce needs.

Remember, the goal is to create a seamless visual experience for your customers. With the right adjustments to your Nano Banana 2 workflow, uneven lighting becomes a solvable problem rather than a permanent obstacle.