Nano Banana 2 Reference Image Preparation Checklist for Consistent Character Design

Creating a character that looks the same in every scene is one of the most challenging aspects of AI image generation. When you use Nano Banana to generate new images, the tool relies heavily on the visual data you provide as input. If your reference images are unclear, poorly lit, or contain conflicting information, the model may struggle to maintain consistency. This checklist outlines the essential steps to prepare your source material before you begin your workflow with Nano Banana 2.

Selecting High-Quality Reference Images

The foundation of consistent character design lies in the quality of the images you upload. Since Nano Banana analyzes these inputs to understand your subject, the clarity of the original file is paramount. You should select images where the character is clearly visible and not obscured by heavy shadows, motion blur, or complex backgrounds that distract from the subject.

For best results, choose at least two distinct views of your character if possible. A front-facing portrait helps define facial structure, while a side profile or full-body shot can clarify clothing details and proportions. Avoid using images where the character is partially cut off by the frame edges, as this forces the model to guess missing geometry. Remember that prompt instructions describe desired outcomes but do not guarantee identity preservation; therefore, the visual evidence in your reference must be strong enough to carry the weight of the request.

It is important to note that while the prompt library offers example prompts that users can copy, these examples serve as starting points. They do not replace the need for high-quality visual references. Always treat any specific prompt example provided in documentation as an untested illustration of syntax rather than a guaranteed formula for success.

Formatting and Technical Constraints

Once you have selected your images, you must ensure they meet the technical requirements of the platform. Nano Banana 2 supports text-to-image and image-to-image workflows, but the efficiency of these processes depends on how you format your inputs. Ensure your files are in standard formats such as JPEG or PNG with sufficient resolution. While the tool does not explicitly list a maximum pixel count in general documentation, extremely large files may slow down processing without adding value to the output.

Be mindful of which version of the tool you are using. Google documents Nano Banana 2 Lite as focused on speed and cost. It is not optimized for multiple reference inputs or multi-turn sequential editing. Do not recommend it for those workflows without explaining this limitation. If your goal is to maintain strict character consistency across many generations, the standard Nano Banana 2 interface is generally more suitable than the Lite version. The Lite variant is designed for quick iterations where perfect fidelity is less critical than rapid turnaround.

When preparing your images, avoid adding watermarks, logos, or unrelated text overlays unless they are part of the intended design. These elements can confuse the model's interpretation of the character's actual features. Keep the background simple or neutral to allow the algorithm to focus on the subject matter.

Executing the Workflow and Evaluating Output

After your images are prepared, you are ready to initiate the generation process. Upload your selected references into the Nano Banana 2 interface and pair them with clear, descriptive prompts. Use the prompt instructions to specify the desired outcome, such as "character walking in a forest" or "portrait in studio lighting." However, remember that prompt instructions do not guarantee identity, label, object, or typography preservation. The AI will attempt to blend your reference data with your textual description, but variations are expected.

To judge the results, compare the generated images against your original reference sheet. Look specifically for changes in eye shape, hair texture, and clothing patterns. If the character looks significantly different, revisit your preparation checklist. Did you include enough reference angles? Was the lighting in the source image too harsh? Sometimes, adjusting the contrast or cropping the image to center the face can improve the model's ability to lock onto specific features.

If the results are inconsistent, try refining your prompt to emphasize specific traits mentioned in the reference, such as "red scarf" or "blue eyes," rather than relying solely on the image upload. For complex projects requiring multiple turns of editing, ensure you are using the correct model tier, as Nano Banana 2 Lite is not optimized for multi-turn sequential editing.

By following this structured approach to reference preparation, you increase the likelihood of achieving a cohesive character design. Start by gathering your best assets, format them correctly, and then engage with the tool thoughtfully. Try Nano Banana to begin your own character creation journey today.

Troubleshooting Common Issues

Even with careful preparation, you may encounter issues where the character drifts from its intended look. If this happens, first verify that you are not using Nano Banana 2 Lite for tasks requiring multiple reference inputs. Next, check if your reference images contain conflicting data, such as a character wearing two different outfits in the same set of uploads. The model may get confused trying to reconcile these differences.

Another common fix involves simplifying the prompt. Overly complex requests can sometimes override the visual cues from your reference images. Try breaking down your request into simpler components. Finally, ensure that the images you uploaded were not corrupted during the transfer process. Re-uploading fresh, clean versions of your source files often resolves subtle rendering errors.