Topaz Photo AI v0.8

When using standalone PhotoAI, Sharpening crashes program right away, I have 12-core Amd cpu and 3090 nvidia.

A bit off topic but “If the Installer can do this in Photo AI for RAW files could it not be incorporated into the stand alone products also”?
Thankyou.

Downloaded fine and installed OK. Opened as standalone and was pleased to see the RAW adjustment menu. Noticed onscreen that the selection of the car ( the Porsche shot) was not 100% , would like to brush adjust. Then processed for screen using auto, and noticed that the noise adjustment was limited to the selected area and not background. Changed the menu to include noise reduction ( auto values) and it now noise adjusted the background and only sharpened the subject ! This time no nasty sharpen of the oof areas or the white car ( woo hoo ) Unlike v7 where the raw conversion was horrid in terms of colour and saturation , this preview looked much better… sadly that is where the good news ends. When processed the file as DNG and opened in Photoshop I was horrified to see it looked way too bright , not at all like preview. Checked again to make sure no RAW adjustment was dialed in and it was all set to zero. File in PS all looked as bad as the v7 RAW conversion , I am using a reference screen with average delta E of .35 ! So for me it is progress, add brush adjust for selection, court martial the RAW team , and feed the whole process loads of uppers to drive it faster.

I’ve been using v0.7 last week in connection with Apple Photos. I was able to round-trip a jpeg from Photos via an older app extension called External Editors. All Topaz apps used to work this way. DeNoise AI and GigaPixel AI still do, Sharpen AI changed around v3.3, where you couldn’t save an image back into the Photos Library. I reverted back to v3.2 and haven’t upgraded.

To be clear, while working on an image in Photos, you can select External Editors from the extensions menu, select DeNoise AI, the app launches, make any edits necessary and then save the changes. When processing concludes, the save window closes, the app quits and the user is returned to Photos in the External Extensions extension where you can select Cancel or Save Changes. Saving returns to the Photos edit module, and all changes are viewable and you can revert to original if desired.

Photo AI v0.8 no longer works like v0.7, saving edits doesn’t register with External Editors. I don’t know where it’s saving the image, the app remains open and doesn’t return the user to Photos.

Now, Photos users must export versions of their images and then re-import, causing duplicate images and extra effort in the workflow. If I can’t round-trip an image from Apple Photos to Photo AI, I will not be a customer.

For the moment, I’ve reverted back to v0.7.

Good update but could Photo AI get a art enhancer setting?

I have an AMD RX580 (4GB) - when resizing, the video card crashes (if I choose the CPU for processing, crashes do NOT occur, bit of course processing is MUCH slower). I have to shut down the computer and restart. I’ve had this problem for at least a few months, with DeNoise, Gigapixel, and Sharpen. Before that time, all was well, the RX580 worked fine.

I posted this issue back in February in community.topazlabs.com:

I’d think it’s MY video card/drivers/computer, but I see other folks are having the same issue.

The problem is intermittent (but often), happens with all the drivers I tried (right up to the latest of course), and is a real problem because not only do I lose my work in the Topaz product, I lose work/tabs/etc, in other programs I have open.

I was hoping Photo AI might have fixed the problem, but it has not.

Thanks…

I like the addition of Raw adjustments, but it i difficult to use since moving a slider cause the model to rerun and the impact of the slider becomes a guess. It’s a little better if you turn off the other features but still lags too much for efficient use.

I also find that PAI cannot recover dark images nearly as well as other programs. There are files that are unusable unless pre-processed.

Raw image exposure in PAI

Exposure adjusted before PAI

2 Likes

Thanks so much for the continued improvements!

I really like the RAW adjustments. I often shoot indoor sports underexposed and the simple brightness correction allows me to see much better what PAI is doing in the other filters. I will still export back to Lightroom for final edits but being able to “make a start” in PAI is great.

I will also repeat my prior requests (hopefully already on your list):

A crop tool as part of the resize filter would be HUGE for me. That would bring all my potential RAW adjustments into one place. Being able to just use PAI to remove noise, sharpen and then crop and resize to give a single DNG that can then be “finished” in Lightroom would really improve my workflow A LOT. Fingers crossed this is where PAI gets to.

Masking to match current functionality in Sharpen AI, and adding the ability to select/de-select faces

Guys

I’m still getting a red cast to the images in Photo AI that I reported began back with the upgrade to V.7.

I pull RAW files into LRC and convert to DNG. When I move them over to Photo AI they all have a distinct red cast, almost looks like a mask over the entire image. The cast does not go away and is saved with the image back to light room. I am attaching the original DNG file from Lightroom with my edits and the processed Photo AI version. I converted both to JPG for uploading.

LRC V11.5
MacBook Pro OS Monterey V12.4


0.8 crashes on Mac M1 pro when plugin from latest Photoshop is used.

I like the idea of PAI and it definitely has improved some of my photos. In that context, here are some comments on PAI 0.8:

  1. To me, the raw development controls are not useful because real raw developers (I use darktable, but the same would apply to LR, Affinity, Adobe Camera Raw, or other raw-development tools) have so much more useful raw development functionality. I would suggest not trying to duplicate that functionality but rather to focus on the functions that PAI is good at, and otherwise just to create as output a dng file improved by the principal functions of PAI but that can otherwise be processed in customers’ normal raw development workflow.

  2. PAI seems to miss faces when the subject is in profile, especially if they have their eyes closed. This locks out face enhancement, which might be useful if unlocked, at least if user-defined masking were possible.

  3. The image quality number does not correlate with anything I know how to see.

Finally, a question. Is the denoise AI function luminosity-aware? That is, does it know that there is usually more noise to be found in the dark areas of the photos and therefore does it automatically apply more denoising in those areas?

Despite these issues, it appears that PAI is shaping up to be a great product. Keep at it!

2 Likes

AIV08 Standalone. Mac M1 32Gb. Latest OS.

What I have noticed is that this product does very well when there is detail in an image, but it struggles when there is no detail – especially noticeable in either the shadows or highlights. I have attached four images downsized to 1800 pixels height to illustrate this.

The first image shows just PhotoAI output file as is (scaled down of course). The second image shows the same PhotoAI output image but uses an extreme curve to emphasize the embedded blotchiness in the out-of-focus shadows. The third images is a composite which exaggerates what is going on in the shadows and highlights by outputting two versions of PhotoAI (normal vs. strong noise reduction) and then differencing them – notice the obvious grid pattern being introduced by PhotoAI as it struggles to make sense of the out-of-focus, grainy, highlights and shadows. The fourth image is the original file with the same extreme curve to emphasize the shadows – notice no blotchiness. Also notice that PhotoAI has completely missed identifying the horse’s mouth near the bit, which is clearly visible here.

Given what I know about ML algorithms this is going to be a tough nut to crack but I think it essential that it be fixed if pros are ever going to embrace this product. Comments welcomed.

We’re also feeling out a new feature related to image quality. Images will be shown an image quality score with a percentage that represents how much your image improved.

@anthony.lawn This feature might be useful if it were visible… :laughing: Someone needs to change the colors for this… Us 1/2 color blind folks can’t see greyed out green text on a grey background…

Tested RAW files from Nikon P1000

Great to be able for the first time to get reasonable colors from RAWs because demosaic itself looks good

Don’t known if you apply automatically a specific profile for each camera ? Would be great to have this profile refine.
It will be cumbersome to move 9 sliders to adjust for every RAW so at least you need to give the capability for the user to store profiles specific to his cameras and to be able to recall it and also at least to keep in memory the last profile used
Of course I hope you will be able to make preview adjustments much faster and to add histograms to help find the good settings.
Please keep going .This is a big plus if we can often avoid to go first through another developper

1 Like

Hi:
First off, I love this product. I do a lot of bird photography and notice I often have to add sharpening when it chooses noise reduction. Hope this is helpful.

Assume since Adobe LR only exports as DNG to access these new features you have to import directly? Is there any workaround from within Lightroom?

Fred

I was trying the On1, painter and PSP plug-in options. The good news is the plug-in shows up and works. The bad news if you are working with a transparent background layer, you lose transparency. This may be expected behavior when passing images between apps.

Used Win installer to update. Installed relatively quickly. No issues w/install.

My sys specs: Win 11 Pro desktop. Auto Processor (I have AMD RX6800 XT & Intel i9 12th gen - don’t know what Auto selects…). Standalone PAI .8 version.

Source = .tiff image (previously tested in earlier releases - 2 guys bird watching in field). Only AutoPilot (A.P.) for this round. Resize - I set to 2x. Can we use fractions?

  • A.P. ‘Subj Detected’ - Yes, partially (it didn’t detect the men when trained to ignore <5-10% of image)

  • A.P. says ‘Subj in Focus’ - The image and subjects look soft to me (NOTE: I assume A.P. is telling me what it sees in the orig. & not what it’s done to fix the image… if so, that’s a - consistent across my tested images - misread of the subjects’ sharpness)

  • A.P. says ‘Image Noise Severe’ - It says that consistently too. I see very little, if any, noise in most images, including this one.

  • Enhance is locked to On - I can’t turn it off, can only change b/wn style of enhance (I tried Natural & Low Res neither is great, but low res was preferable of the two).

I checked Split View & Full View (it’s great I was able to use Spacebar to toggle b/wn Preview & ORIG in that view!). Labelled snips are below…

I don’t see any Quality Score — Is that only for Raw images? Oh wait, is that “100%” in box next to Zoom the Quality Score? If so, I thought it was the Zoom amt. Wouldn’t a Quality Rating be better up under all the Assessment Parameters under the AutoPilot area (& labelled). If that is the quality score, I don’t agree with that quality rating based on the factors noted below…

Observations/opinions re: the processed preview for this image:

  • Processed preview is bad in the following ways:

  • Shadow lines in clothing look like computer-generated graphics illustrations (not like a photograph).

  • Likewise the Hair, Ears, Hat all look more like outlined graphics illustrations. This is a photo that is intended to look photographic.
    -White halos have been added around all the shadows on the fabric. This is puzzling to me - there is no “Sharpen” setting added & that’s what I would expect that type of artifact to result from.

  • The Background (landscape) is sharpened disturbingly & is distracting - it’s become over-sharpened squiggles. Looks like brushstrokes. Which it isn’nt. I didn’t use any Studio Impression or any artsy filters on this.

  • Preview looks like it’s been stripped of some saturation or washed out relative to ORIG

  • Other ops (ongoing…) concern: The UI still opens to centered on my monitor display when launched from Taskbar.

Labelled Snips:

That’s a cool pic!

Ps 2022 (rel. 23.5) PAI .8 Plugin launched & ran.

In well-exposed image, all 16 faces in a group shot were recognized. In a separate group shot (not well exposed plus included mix of dark-skinned people and ppl whose bball caps cast shadows on their faces) about half the faces were detected - those not in shade or not with darker skin were detected.

For the image with all faces detected only about half of the people were detected as subjects - the rest, not).

Same problem with the Ps (2022) Plugin that recurs … the PAI UI launches to centered over my Ps workspace.

Same ops consids./concerns as above for standalone.

Hello, according to my tests, I have currently obtained better results with PAI working with NON RAW files. With RAW files PAI still adding a reddish hue and extra contrast and even some artifacts. The basic controls are appreciated but they don’t seem to help much at the moment. On the other hand I’ve also noticed that it eventually blurs certain areas (possibly those with very fine textures). I’m attaching images of my results. (The RAW file is of course at disposal to the Topaz PAI team.)