There is a severe bug in the image processing pipeline when batch processing Nikon NEF files.
When applying mild RAW Denoise in combination with the standard Denoise filter, Topaz introduces heavy rainbow/chroma color artifacts and horizontal banding on fine fabric textures (like a polo shirt). Additionally, a distinct square tiling/grid pattern is visible on the fabric (suggesting a blending issue between image processing tiles).
The community has reported similar color band artifacts since April 2023 (e.g., thread “Artifact on color band after denoise” in Photo AI Bugs and Issues). While the legacy DeNoise AI handled these patterns correctly, Photo AI still fails here.
Exact Settings Used (applied globally to the full image, using the ‘All’ selection):
RAW Denoise: Normal, Strength 20, Minor deblur 10
Denoise: Normal, Strength 10, Original detail 0
Adjust Lighting: Model V3, Exposure -0.17, Highlight -0.17, Shadow -0.34
Sharpen: Standard, Strength 22, Opacity 44
Changing the secondary Denoise model to MAX eliminates the color artifacts, but local processing speed becomes unworkably slow (even on high-end hardware like an AMD Ryzen 9 9950 X3D and NVIDIA RTX 5080 system). Disabling the second Denoise layer removes the artifact, but for batch workflows (>500 images), manual intervention or per-photo mask adjustments are not viable.
This image is not low-light enough to require Denoise. There is no need to add a Denoise model here other than RAW Denoise to remove the RAW noise from the original file.
The artifacts you’re seeing could be caused by overprocessing, or potentially a processing issue with the graphics card. A few things to make sure of:
Only apply Denoise to images that actually have low-light noise. It is not needed on this particular image.
Make sure your AI Processor is set to Auto under Edit > Preferences > General > AI Processor. This allows Topaz Photo to use your machine’s full capabilities rather than forcing a specific processor.
Make sure your graphics card drivers are fully up to date.
For batch processing, you can process any darker images that actually need Denoise separately, or you can also use Auto-Pilot, which would analyze each image and would not add Denoise to an image like this when it isn’t needed. USing enhancements that are not needed will produce artifacts.
Let us know if you have any questions after testing the steps above! You can also DM-me the file if you want us to test what best enhancement combination is to use here and to see if we can replicate as well the artifact on our side.
Thanks for the clarification regarding RAW Denoise vs. standard Denoise on well-lit images. I understand that applying standard Denoise on a low-ISO photo is technically redundant.
However, I still want to point out a fundamental issue with how the algorithm behaves in these scenarios:
Artifacts vs. Overprocessing: Unnecessary noise reduction should at most result in a softer or over-smoothed image. It should never generate rainbow/chroma color artifacts or grid-like banding across fine textures.
Tile-Blending Issue: The distinct block pattern strongly suggests a tiling/patch-based processing issue where the AI fails to properly blend overlapping boundaries on repetitive patterns (like fabric piqué).
Even if an enhancement isn’t strictly necessary, applying it should not cause mathematical rendering errors or blocky artifacts in the image pipeline.
I will stick to using solely RAW Denoise for this batch to move forward, but I wanted to share this insight so your dev team can look into the tile-blending behavior on fine textures.
[Perhaps I’m a bit overly critical, but in the end, we both benefit from a flawless and reliable tool!]
Thanks for the follow-up. I can confirm the following:
System Requirements: My system far exceeds the maximum requirements (AMD Ryzen 9 9950X3D, NVIDIA RTX 5080, 64GB RAM).
GPU Drivers: Cleaned and updated to the latest Studio drivers.
AI Processor Setting: Set to GPU (as previously recommended for high-end setups), but tested on Auto with the exact same result.
Due to privacy and copyright considerations, I do not wish to share the original RAW file publicly or upload it directly to the forum environment.
Please let me know how or where I can securely send/upload the file directly to your engineering team (e.g. via a private support ticket link or a secure upload location).
