Hi-Fi as you chose will do the least amount of detail smoothing or hallucination than the others with the goal of preserving the fidelity of the original and only adding hi-rez details to the existing lo-rez pixels in the source, but you have to turn down suppress noise to zero. The auto-pilot makes very poor decisions and does not seem to be trained on when pictures will benefit from moving the sliders up.
While Standard will give you a more detailed image, when you zoom in you will find it hallucinates unrealistic details. It also has not been trained on what is natural texture segment noise versus what is camera sensor noise and is more likely than Hi-Fi to give you that cartoon smoothing.
There has been a long standing complaint that with nature photos like this there will remain patches of blur and detail even within the same texture segment no matter how much you play with ai model and parameters, but turning the parameters way down minimizes this a lot.
I feel like the issue here is using denoise function (see icons in on the bottom left corner of image list’s preview - green dot, resize, denoise). Hi-fi should not alter the image like this huge. I do not have the image just guess.
Did not see that but likely. Why was it changed to do enhance before denoise/deblur AI as those other AI are about fixing issues in the blurry/noisy source sensors - they are not trained on artifacts from the poorer quality enhance sliders for blur/noise. So it seems to be more of a revised auto flow issue, was denoise/deblur AI followed by a Hi-Fi without parameters so that it would preserve whatever the prior AI did and only add hi-rez over the lo-rez source.
Confusing new UI but I have not updated until I hear seats and saves bugs are truly fixed. With the old UI the sidebar was stacked in processing order but maybe it is just slid down to the upscaling step in the pic.
But I would then agree, it is not likely that the denoise AI would perform well with lo-rez spotty fawns surrounded by lo-rez brambles especially a .very compressed .jpg - none of that noise is from sensor noise.
I think enhance only without any of the other AI, with no suppress noise and remove blur parameter with the fix compression slider turned up. Hif-fi. Standard hallucination will for sure happen with thickets combined with more painted grass streaks. It might do well with the fawns though but I would expect blurry patches as they are within the brush.
With a very small image like this, the output result can vary greatly based on the starting image quality.
If the image quality is very low, then it will be challenging to improve.
You will likely get the best result from upscaling only, with no other filters on. I would highly encourage testing the slider settings as well, such as decreasing Suppress Noise.