Thank you for your correction. I have edited my post.
I meant “Those images are straight out of camera without any post-processing afterward my me.”
Those are old photos, I don’t have raw on those images anymore, otherwise those noise can be removed effectively by DeepPRIME or DeepPrimeXD.
Denoise Strong V1 should stay with TPAI for high ISO images.
1 Like
PAI rel. 1.3.7 - Noise Models. Win 11. Ps Plugin.
Normal v. 2 totally mutated the teeth and eye of this .jpg image. Looks like the person is winking now!!! Yuck!
PAI turned on denoise normal v. 2 w/out me doing do so. A.P. selected it.
That’s not even taking into account the rest of the processing made the image blurrier than the Before image…
Snips of the BEFORE & AFTER Normal v2 below - labelled in filenames.
Working from Photoshop layer. When autopilot selects subject it picks part of the background as well. If it selects a noise reduction and I want to select some sharpening nothing changes in the pre-view when moving the slider or selecting lens, motion of strong. Normal and strong noise reduction seem to work but V2 on both are not changing with sliders.
The V2 version works great, but for some reason in PhotoAI, even at 900 ISO which needed some cleanup with noise, it was not automatically chosen. I had to manually click on it which made it much better.
I’ve tested more images again today - while working on experiments for a new YouTube Ps video.
Consistently, Normal v2 over-sharpens and creates white halos. And, that’s what PAI is using by default now. And, also consistently, Strong v2 softens my entire image(s).
2 Likes
version 2 for me
The new noise reduction works quite well. How ever for some images I got better results with normal vs. normal v2. I need to test a bit more to be sure.
Based on what I keep experiencing, I hope the current Normal & Strong (v1) models will continue to be inherent to the PAI product.
And, the v2 models included along with them - for those who seem to prefer them. Perhaps some very different type of images than the Canon, Sony, Olympus, Panasonic, Samsung (& older phone brands) and Fuji images I work with respond well to the v2 models.
It would be disappointing to stop being able to use a Topaz Labs product after over 13 yrs of use with my photography (the classics being my faves!) & over a year testing, observing, reporting, etc with the PAI product…
The old models should be optional in the settings.
Like it was before with the sharpen models.
I’d like to turn off everything i don’t need.
1 Like
I was under the impression (from something I read on the public forum) that what is being considered is to replace the v1 models with the v2 models, depending on feedback.
That’s what I am responding to. I am not having success (with the exception of a single rare image) with the v2 models. If that’s all that would be supported going forward, I would be SoL (a NY metro US acronym).
You probably know, from my usual posts, I work with settings manually & don’t rely on auto anything (all the Auto stuff is off in my Prefs & I don’t use all sliders just b/c they’re there).
I cannot fix the the image issues I’m experiencing with the v2 models. So, would not want the v2 models to replace or supplant the v1 models as the only processing option in PAI.
PAI cannot be a 1st step, in my always Ps workflow, if it wrecks (rather than positively primes) my images before other ‘creative’ workflow steps. But, perhaps I’m unique in having that concern/requirement.
TPAI cannot currently be the first step, the way the models work now, it only works at the end of the workflow as the last step.
I have never used them in any other way.
Currently I’m trying to make a deal, I hope it works, I would then possibly test a piece of hardware and I thought then I also show how I work with TPAI.
1 Like
But there are exceptions.
I just found out that there can be problems when you process an image in the RAW converter (80% done), then edit and denoise.
It can be that the noise changes in such a way that TPAI can’t handle it to 100%, because it has changed so much through the processing that it becomes a type of noise that TPAI doesn’t know.
With these changed types of noise the old stong model can handle it better.
In that case -->Raw converter edit → denoising → edit in PS etc.
The old Nomal model always creates rainbow colors around strong contrast edges.
2 Likes
Yes, wrt what happens to my photos. But, I thought it was positioned as a pre-processor. So, that’s how I’ve been trying to use it. At least in the sense that I, personally, didn’t do any other processing to an image straight from camera. As we know, a jpg is already processed by the camera.
Mostly, I’ll launch a raw image directly into ACR, then use the PAI plugin on whatever form of that image opens into Ps & then into PAI…whether .tif or whatever pixel format w/out me doing any Ps manipulations 1st.
I use it at the end of the workflow. In my experience, every new update tendos to be overdone, and especially v2 of noise reduction and sharpen. Waaay overdone and unrealistic results. Original is much better, and DenoiseAI and Sharpen AI better yet. Thanks.
1 Like
For me Normal V2: 1. misses parts of the image leaving blotches or blocks of the image that are easily discernable. 2. Does not seem as easily controlled even though there’s a lot more “room” in the sliders. I have not tried Strong V2. I’m assuming the similar results to be expected.
On my windows computer, I’ve seen an error where it tries to use v2, before telling me something about my graphics card doesn’t support it (sorry, didn’t save the message). I then manually select the v1 model. Not sure if this was a bug or intended behavior but I’m hoping that if a model isn’t supported by the computer, then we can use a preference to pick the model to use, instead of it trying and then presenting an error before I have to manually change the model. The experience should be more seamless. Obviously, I hope my computer will support v2 final version. However, if not, I don’t think v1 should be dropped.
I agree - keep both! I have found that in bird and squirrel images the V2 makes the feathers/fur too soft. For me normal works better in those instances.
2 Likes
I find the new models to be troublesome and I find myself using the old models 90% of the time even though I try the new ones first with a range of settings. I see the new models doing well on parts of the image but leaving areas looking unprocessed and even worse than before. Clothes and around the edges of faces are especially bad. Also there are a lot of hot pixels. I realize this is in Beta but should the new models be implemented and the older models removed, sadly I would not have any use for the app. My images are very noisy from dark venues so maybe they are unusually difficult but Denoise and Photo AI have handled them pretty well up til now but the new models are nowhere near good enough, at least for me.
1 Like
Keep both, simply because on some pics v1 works best, on others v2 gives better results. Strong v2 definitely has some problems creating artifacts where V1 doesn’t.
1 Like
