It's not ideal, but it's the best I have for now before darktable devs can integrate nind-denoise (or other denoisers).Flatpak build available in: x86-64 and AArch64 darktable-cli would export the first stage with minimum processing (demosaic, WB) into a 32-bit TIFF, which is fed into nind-denoise and output as another TIFF, then darktable-cli would apply the second stage of the remaining operations (lens correction, rotate/crop, contrast, heal. My current work-around for darktable is to edit the RAW normally, but instead of exporting directly from darktable at the end, I run a Python script that splits the XMP history stack into two stages. I guess there's no way around it as the denoising process is quite time-consuming, doesn't allow real-time editing. And if I need to change brightness or WB for example, rinse and repeat. I think it's quite cumbersome currently to integrate these neural-network denoisers into a typical photography software, users have to convert the RAW into an intermediate format to feed into these denoisers, then import the denoised images back into the editors. Thus, I always cap AutoISO at 800 and bump up brightness in post instead, more headroom for highlights without penalty for noise. Sorry, I could have taken the shots individually and bumped up the ISO accordingly, but I intended to use these as training samples for nind-denoise as well, so I AE bracketed it instead to minimize alignment later.īesides, my X-T2 has dual-gain that switches from DR-optimized to sensitivty-optimized at ISO800. That said, it would be far more useful for Topaz to add some basic raw conversion features This is case where some straight high ISO images would be useful, as I wouldn't have to apply exposure compensation to them. Unfortunately Topaz PhotoAI doesn't yet allow for exposure adjustments of raw files.

Darktable 32 bit