Normal denoising is useless as it forgets part of the image, it's not consitent

[vincent.lugnier (VincentR5)]

Hi pretty much never use the normal setting because I can’t as it produces areas with no denoising, and others with denoising. So you hav patch of areas with full noise

[preston.werner]

Hi @vincent.lugnier, this is expected behavior for the denoising algorithm – some areas of an image are harder to denoise than others, so depending on how aggressive your slider’s settings are you may find that some areas are not denoised.

I may be misunderstanding, though, are you saying that you never use Autopilot since it produces these results so you always have to manually adjust it?

If so, this is something I can let our developers know about as they continue to train and refine the autopilot adjustments.

Thanks!

[vincent.lugnier (VincentR5)]

I’m saying that since with the normal setting I ALWAYS get un-denoised areas, I ALWAYS uses the strong setting, that NEVER leaves some areas un-denoized (but unfortunately strong is often too much, even at 1 setting).

I think that your expected behavior doesn’t make sens at all, if you have patches of un-denoised areas amongst smooth ones they show up even more, and that’s even worse than leaving the noise.

And as I said, it doesn’t occur with the strong model

[Lingyu]

This is a known issue with the Remove Noise Normal and Strong models.

We are currently almost done training a new version for these models to improve the performance and are releasing a beta to test the results.

The updated models will be better at removing noise and the Autopilot settings should produce a better result the first time around.

Please let me know if you have any questions about this. I expect the new models to be released in a few weeks.