Comparing Blender's Denoising Methods: Which One is the Best?

Table of Contents

1. Introduction
2. Comparison of Blender Denoising Methods
   • 2.1 Open Image Denoise (OID)
     • 2.1.1 Performance at Low Sample Count
     • 2.1.2 Performance at High Sample Count
   • 2.2 OptiX
     • 2.2.1 Performance at Low Sample Count
     • 2.2.2 Performance at High Sample Count
   • 2.3 Topaz AI Denoiser
     • 2.3.1 Performance at Low Sample Count
     • 2.3.2 Performance at High Sample Count
3. Recommendations for Animation Rendering
4. Saving Original Image vs Denoised Image
5. Conclusion

Comparison of Blender Denoising Methods

Blender offers several denoising methods to enhance the quality of rendered images. In this article, we will compare three popular denoising methods: Open Image Denoise (OID), OptiX, and Topaz AI Denoiser. We will examine their performance at both low and high sample counts, taking into account the quality of noise reduction and preservation of details in the image.

Open Image Denoise (OID)

Performance at Low Sample Count

When using Open Image Denoise (OID) with a low sample count of 15 and a resolution of 1080p, the results are impressive. OID effectively reduces noise in the image, especially in areas with subsurface scattering. The skin appears smooth and the cloth folds on the character's top and shorts are preserved with minimal noise. However, there is a slight downside - the specular highlights on the shorts appear smoother than desired. Additionally, patches of red and blue can be seen on the torso, which were Present as noise in the original image.

Performance at High Sample Count

At a higher sample count of 1000 and a resolution of 1080p, OID continues to deliver impressive results. The image quality significantly improves, with fewer compression artifacts compared to low sample counts. The cloth threading detail on the top is clearly visible, indicating OID's effectiveness in preserving fine details. However, the specular highlights on the shorts are overemphasized, introducing lighting that does not exist in the source image. Despite this, OID outperforms OptiX and Topaz AI Denoiser in terms of noise reduction and detail preservation.

OptiX

Performance at Low Sample Count

OptiX, another denoising method in Blender, falls short in comparison to OID. At a low sample count of 15 and a resolution of 1080p, the image appears heavily compressed. Large patches of red and blue are visible on the skin, which are actually present in the source image, but get distorted by OptiX. The top of the shorts loses its original pattern and looks smudged.

Performance at High Sample Count

With increased samples at 1000 and a resolution of 1080p, OptiX shows some improvement, but still lacks the quality and detail preservation offered by OID. The image appears less compressed, but the cloth folds on the clothing are lost. OptiX struggles to retain threading detail and introduces bluriness. In comparison to OID and Topaz AI Denoiser, OptiX fails to deliver satisfactory results.

Topaz AI Denoiser

Performance at Low Sample Count

Topaz AI Denoiser, while a separate denoising solution, is also compared to OID and OptiX. At a low sample count of 15 and a resolution of 1080p, Topaz AI Denoiser performs poorly. The skin looks terrible despite efforts to reduce noise, and the shorts lose all their folds. The results are far from desirable, with OptiX and OID outclassing Topaz AI Denoiser in terms of noise reduction and preservation of details.

Performance at High Sample Count

Even with higher samples at 1000 and a resolution of 1080p, Topaz AI Denoiser fails to overcome the limitations observed at low sample counts. While there is some improvement in image quality, Topaz AI Denoiser still cannot effectively reduce noise in the skin, causing it to appear unnatural. The denoising method also removes surface details in the gloves, further compromising the image quality.

Recommendations for Animation Rendering

Based on our analysis, it is advisable to use denoising techniques when rendering animations, unless the scene is exceptionally simple or the rendering can be offloaded to a rendering farm. By using denoising, you can achieve satisfactory results with a lower sample count, reducing the overall rendering time. Open Image Denoise (OID) proves to be the most effective denoising method, providing excellent noise reduction and detail preservation.

Saving Original Image vs Denoised Image

To address potential loss of detail caused by denoising, it is recommended to save both the original image and the denoised image. This can be achieved through the use of the blender compositor and connecting the render outputs to a denoise node. By saving both images, you have the flexibility to choose between the original image, which retains 100% of the detail but may have some noise, and the denoised image, which provides noise reduction but may smooth over certain areas.

Conclusion

In conclusion, the choice of denoising method in Blender plays a crucial role in achieving high-quality rendered images. Open Image Denoise (OID) demonstrates superior performance in noise reduction and detail preservation compared to OptiX and Topaz AI Denoiser. For animation rendering, denoising is highly recommended to optimize rendering time while still obtaining satisfactory results. Remember to save both the original and denoised images to maintain flexibility in post-processing.