Question about autopilot values

I see that you added weighing (very weak to strong) in 2.0.0
For me that seems to have very little effect. Even if I set it to strong, I still get very low autopilot settings (e.g. minor denoise 1, minor deblur 7 for std-model).

I often end up using the old 1.3.2 version of Photo AI, because it chooses more desirable autopilot values, which preserve much more details of images (e.g. minor denoise 5, minor deblur 55).
1.3.3 and higher come back with less details and a more plastic look, which is very sad.
I wish there was a way to re-equip 2.x with the autopilot detection values of 1.3.2, then I wouldn’t have to switch back and forth between two versions. I was hoping that those new weighing sliders would do exactly that…

P.S. You left out a slider for “Fix compression” in the new preferences. Could you integrate that? Often a very low value like 1 or 2 produces good results. But right now one can’t specify that preference.

Thanks in advance!


@PWC Thanks for sharing your insights! I’ve shared this with our developers so they receive this feedback. Can you share some photos where you’re finding the current Auto detection models to be inadequate?

Hi @tim.he - you can use this one, posted by another user in the 2.0.1 thread a few days ago.
It’s as good as any example:

PhotoAI 2.0.1
AutoPilot suggests supress noise 16, remove blur 19
which results in a lot of loss in detail (e.g. the red car on the poster) and produces horizontal lines (that look like leather on arms and face + on the curtain). The computer monitor in the mirror background also looks very strange.

PhotoAI 1.5.4
The latest 1.x version also detects similar values, supress noise 13, remove blur 16.
Resulting in an almost identical image (slightly better than 2.0.1 if you ask me):

PhotoAI 1.3.2
does the best job with detected values 73,37
which results in a lot of additional detail (e.g. red car, left eye from camera’s perspective) and almost no leather lines (arms and face). Also the computer monitor looks pretty good there. The curtain on the right side also looks very good without any horizontal lines. The paper ball in the shelve also looks much sharper there.

Here also a more detailed view (order: 2.0.1, 1.5.4, 1.3.2):

All default settings, except subject detection (that was completely off). No manual tweaking was done.

This happens to me with many images - mostly low light scenes and low res (images around the size of 1200x675). I wish I could somehow tell 2.x to detect it like 1.3.2 did, then I wouldn’t need to use 1.3.2 any more (my hope was, that the new autopilot preferences with the weighing values would do exactly that; but they don’t help there).


The 2001 me is famous :slight_smile:

One additional question: is there any way I can force PhotoAI to always apply sharpen to every image? (via preferences) I tried setting the “Enabled when detected blur is…” with the lowest setting “low soft and above” in the prefs, but it’s never hit and I’ve tried lots of pictures that had motion or lens blur.

I would like to force it, since I already know which pictures need that (I named them with a _blurred preference) and I could run them all through the CLI. But that would only work if there was some way to “Always” run a model [the same option would also be useful for the other filters like Remove noise - maybe you could add the value “always” as an option for the “Enable when detected…” dropdown]

Thanks again.


@tim.he is that analysis (a few posts up) with the sample image and comparison to three different versions of Photo AI helpful for you and your team?

Yes, very much so. It helps us see how people are actually using the app. We have our own assumptions but they’re not always right. It also helps us in our own testing too.