Nano Banana 2 Image-to-Image: Cleaning Scratches and Dust from Digital Paintings

When scanning physical artwork or working with older digital files, imperfections like dust specks, paper texture noise, and surface scratches can detract from the final piece. The goal of digital cleanup is not to erase the artist's intent but to refine the canvas. Nano Banana 2 offers an image-to-image workflow designed to handle these specific restoration tasks. By leveraging its prompt library and understanding the underlying model capabilities, users can achieve a cleaner result without flattening the texture of the original strokes.

It is important to clarify that Nano Banana refers to the AI image generation and editing tool in this context. It is not a skincare brand, bottle, jar, or physical subject. This guide focuses on using the tool to clean up scanned artwork or digital files, specifically targeting the removal of imperfections while maintaining the integrity of the artist's hand.

Structuring Your Prompt for Cleanup

The effectiveness of the cleanup process relies heavily on how you construct your prompt instructions. Unlike text-to-image generation where you describe a scene from scratch, image-to-image prompts in Nano Banana 2 must describe the desired outcome relative to the input image. The prompt instructions describe desired outcomes; they do not guarantee identity, label, object, or typography preservation. Therefore, clarity is key when asking the model to remove specific artifacts.

To target scratches and dust effectively, your prompt should explicitly state what needs to be removed and what must remain. A strong structure involves three components: the action (remove), the target (scratches/dust), and the constraint (preserve brushstrokes). For example, instead of simply saying "clean image," a more effective instruction would be "remove all dust particles and surface scratches while keeping the original oil paint texture and brushstroke direction intact." This approach guides the model to focus on the noise rather than reinterpreting the entire composition.

Users can access the prompt library within the interface to find example prompts that serve as starting points. These examples are untested prompt examples intended to inspire your own variations. You can copy them or take them into the generator to adapt for your specific artwork. Remember that Google describes Nano Banana 2 as Gemini 3.1 Flash Image, which is distinct from other models like Nano Banana Pro or Nano Banana 2 Lite. Each model has different strengths, so selecting the right one for your cleanup task is crucial.

Step-by-Step Restoration Workflow

Executing a successful cleanup requires following a logical sequence of steps within the Nano Banana 2 interface. This ensures that the model receives the correct context and constraints to produce high-quality results.

1. Upload Your Source Image: Begin by uploading the scanned artwork or digital file containing the scratches and dust. Ensure the image is clear enough for the model to distinguish between the art and the damage.
2. Select the Correct Model: Choose Nano Banana 2 (Gemini 3.1 Flash Image) for this task. Avoid using Nano Banana 2 Lite for this workflow unless speed is the only priority, as it is focused on speed and cost and is not optimized for multiple reference inputs or multi-turn sequential editing. Using the Lite version for complex cleanup might yield inconsistent results regarding detail preservation.
3. Enter the Cleanup Prompt: Input your structured prompt into the text field. Use the example structure discussed earlier: "Remove dust and scratches, preserve brushstrokes."
4. Adjust Strength Settings: If available, adjust the image strength or denoising level. A moderate setting usually works best to remove small imperfections without altering the underlying colors or shapes too drastically.
5. Generate and Review: Click the generate button to create the output. Compare the result with the original to ensure the scratches are gone and the texture remains natural.

For those looking to explore the full capabilities of the platform, Try Nano Banana.

Evaluating Results and Troubleshooting

Judging the success of your cleanup operation involves checking two main criteria: artifact removal and texture fidelity. Did the dust disappear? Yes. Do the brushstrokes still look like brushstrokes, or did the AI smooth them out into a flat wash? If the latter occurred, the prompt may have been too vague, or the model strength was too high.

If the results are unsatisfactory, consider the following fixes. First, refine your prompt to be more specific about the texture. Instead of just "preserve texture," try "maintain visible impasto and stroke edges." Second, if the model is hallucinating new elements where there were none, lower the influence of the prompt or increase the weight of the original image if the interface allows. Third, ensure you are not using Nano Banana 2 Lite for this specific type of detailed work, as its limitations regarding multi-turn editing could prevent fine-tuning the cleanup in subsequent passes.

Finally, remember that prompt instructions do not guarantee identity preservation. While the goal is to keep the painting looking like itself, the AI interprets the request based on its training data. Always review the generated images carefully before saving them for final use. By combining precise prompting with the right model selection, you can effectively restore your digital paintings to their pristine state.