Pixelmator Pro gets a magical, AI‑powered Deband feature
pixelmator.com
pixelmator.com
Pixelmator has been my go to graphic design application on my M1 Mac for the past year now and it feels good to cut the Adobe cord.
Edit: I don’t shoot raw now. I’ve found it unnecessary. I’m not a professional :)
(I haven’t done serious event photography since before Aperture was deprecated, though.)
Same, after spending many weeks evaluating commercial and open-source alternatives. (For Lightroom users who can't imagine Apple Photos replacing it, for advanced editing you'll be using Photos in concert with other apps via Photos' extensions mechanism and/or "Edit With".)
Something like Capture One is much better suited as a replacement for Lightroom. I wish that Apple had a Lightroom alternative, but they got rid of Aperture, so... they're probably not interested in competing in this space again any time soon.
Apple Photos has great built-in RAW support, so no processing is needed. If I want to use a different RAW processor, that's straightforward as well.
> I wish that Apple had a Lightroom alternative, but they got rid of Aperture, so…
Check out RAW Power¹ (macOS and iOS), created by the former lead of the Aperture and iPhoto engineering teams. It's able to use your Apple Photos library directly.
Never heard of people using Photos for RAWs, would love that too.
No. There's no more brand specific versions at half the price. You now pull the full price - even with a Fuji or Nikon.
The subscription is roughly $19/month (if you pre pay annually - else it's $29/month) which is more than Adobe's offering. And the perpetual licenses got shitcanned recently (I bought one on Black Friday) - when you buy a perpetual license you won't get any minor updates for that version anymore. You buy C1 23.1 and they release 23.2? Well, bad luck. But there's a loyalty program where you can buy a paid update. Hehe.
Oh, and their iPad version is an extra subscription at $5/mo.
As a company I detest C1 more than Adobe. But their processing is so great. When I import my RAWs into C1 I don't have to do much with them. They look astounding out of the box.
That's not true for the full version.
Regardless, I tried Capture One and Lightroom Classic side by side, and I liked Capture One a lot better. Even though it is more expensive for people who intend to upgrade regularly, it is worth every penny in my opinion, but I also think a lot of people will be able to get along fine on the freeware license. Capture One handled importing new batches of photos surprisingly better than Lightroom Classic, which actually locked up for long periods of time, and all sorts of other operations seemed faster, as if the underlying engine was not ancient. I don't remember all the details as I did this comparison about a year ago. I did not consider Lightroom CC to be a valid option for various reasons.
I also strongly appreciate being able to have a perpetual license, instead of being forced to subscribe for the rest of my life if I want to be able to continue accessing all of my RAWs with the adjustments I make, and not just the exported photos.
The one thing I disliked about both Capture One and Lightroom at the time is that neither supported proper, end-to-end HDR workflows. I have HDR displays. The RAW files contain a stunning amount of dynamic range. Even the common iPhone captures HDR photos (and I don't just mean HDR processing that gets stuffed into an 8-bit JPEG, I mean an actual HDR photo in 10-bit HEIF with an HDR curve). I want to be able to edit and view my RAW photos with extended dynamic range, not be limited to viewing an 8-bit SDR image while editing, and not be limited to exporting crappy 8-bit JPEGs that are missing so much dynamic range (or gargantuan TIFF files that have limited use cases). Adobe has started adding some HDR stuff this year, and that's the only thing Adobe offers that Capture One lacks that I'm even remotely interested in.
Incidentally, I'm also upset that Chrome removed JPEG XL, which promises to be a common format with proper HDR support.
That’s not an “even” thing; the iPhone has the /only/ end to end HDR still photo pipeline.
Other cameras can do HDR video but not stills. The newest Sony and Canon cameras do support an HDR still colorspace, but not sure what image formats work with it.
Regardless, the RAW files have plenty of range that I would expect modern RAW processing software to be able to emit HDR photos from them.
My use of the word “even” was really meant to be a reprimand of the camera industry. If even an iPhone can do this, a “proper” camera should be able to do it even better. Extended dynamic range is incredibly important to make images photorealistic. Being limited to SDR should be pretty embarrassing for any photo-related company in 2022. But, that’s just like my opinion.
I suggest Lightroom 6 for those who has older camera, otherwise discounted standalone version of Capture One (often goes in discounts). DAM feature is great in Lightroom; but this can be resolved by using some other app Can not wait to ditch Lightroom though. C1 is great for tethered sessions.
Rethinking DAM, i think i need just a good filemanager, i am already organizing my folder by myself.
If you are working in professional environment as a photographer, Capture One is a quite gold standard for photographer. In anycase, retouching is done in Photoshop.
Yes, I have many years' worth of (over 10,000) RAW photos from DSLRs and compact cameras.
> Editing in external applications also isn't very seamless. Just wondering if I'm missing something…
I don't think you are — external editing is just not as elegant as an all-Lightroom workflow. That said, RAW Power (created by the former lead of the Aperture and iPhoto engineering teams) is quite nice and supports working directly with the Apple Photos library.
I'd love to know if there is a good solution for saving time when you have hundreds of photos to process.
I’m consistently impressed with the range and quality of features they add - it got video support only the other day. Won’t be long before it’s a motion design tool as well.
AI upscaling may not be as good as something like Topaz Gigapixel or RealESRGAN, but it’s damn handy to be able to enlarge a layer without worrying about image quality.
For each source, they must develop a "grain profile" (or, likely multiple profiles) so they're applying the right grain ""effect" when re-graining?
And on the re-graining: is this being done on the client side??? Like, is my Roku adding back the grain to my Paul Thomas Anderson films?
(Note: I'm only 5 min through the video, so these may be answered)
However, it doesn’t seem extremely great and tends to smooth out things it shouldn’t. I have it turned off. It seems like debanding in realtime is still too big of a task even for the most expensive TVs.
Third party desktop software could also do this were it not for HDCP. I'm pretty sure any attempt to front-run the stream with a full-screen image processor will blacken the display.
The solution is a drop in decoder replacement. Heck it might already exist, it's a pretty obvious, easy and direct application.
https://mobile.twitter.com/jonsneyers/status/157337162413241...
Netflix has strong history of optimizing their streams. I bet they will be first to deploy.
It’s possible they’re viewing in a too bright setup and are seeing things they weren’t supposed to.
I love it so much I recently decided to make pxdlib [0] to automate fiddly bits like data entry or repositioning layers. It's surprisingly fun to reverse-engineer [1]; the `.pxd` format is a bizarre maze, with common properties some 6 encodings deep!
[0]: https://github.com/yunruse/pxdlib [1]: https://github.com/yunruse/pxdlib/tree/production/docs/pxd
1. Heavily compress image
2. Use neural net upscaler to upscale back to original size/quality
3. Record the diff between the real uncompressed image, and the upscaled image. This should be small if your neural net is good.
4. Store the compressed image + the diff.
This is a lossless image compression algorithm. Part of the trick is that the neural net size is not included in the compressed file size, because that is amortized across many different files you'll be compressing/decompressing.
It's not super portable, though, because whoever is opening your file needs identical neural net weights to get the identical output to apply the diff to. It wouldn't surprise me if eventually one of these neural nets gets baked into a standard somewhere, and every browser ships with that network baked in. Call it NeuralPNG or whatever.
Perhaps these modern day generative programs(sigh, ok ok, "AI") is our generations solution to fractal encoding and the problem with scaling the graduate student algorithm.
"Well we don't exactly know how to solve it but we took 60 Gb of linear algebra and hit it with a stick until it figured it out, no idea how it works but it appears to do the job"
It's likely it would work much better on modern hardware, but it was heavily IP-encumbered back in the day which caused almost everybody working on image compression algorithms to ignore it.
The patents have probably all expired by now so it might be worth another look.
It was also horrifically expensive to license from what I remember which meant that almost no-one used it.
There is really no solution to this because even we couldn't tell this is wrong without prior knowledge of the left side.
https://arxiv.org/pdf/2003.03808.pdf
https://www.theverge.com/21298762/face-depixelizer-ai-machin...
Compression/decompression shouldn't be used for subtle details.
We've seen problems with it before, with scanners[0].
[0]https://www.pcworld.com/article/453236/xerox-scanners-found-...
It'd learn what a scene looked like and then switch to send a really low quality stream to the other person to reconstruct or something like that.
As with this release, Pixelmator Pro has prioritized ML-powered tools and has been an early adopter of new Mac technologies like utilizing the neural cores on M1/M2 processors.
Affinity Photo is in some ways a more traditional photo editor, and its biggest selling point may be its integration with the other members of the Affinity suite, Affinity Designer and Affinity Publisher.
“Professional image editing tools that anyone can use.”
“Designed and built exclusively for Mac”
It’s more likely that people who know image processing are so rare that even an image editor can’t find someone to write a filter using actual math. As usual, all the actual good people are amateurs posting on doom9.
Debanding has been a thing for many years now, for frameservers (Avisynth, Vapoursynth), video players (madvr, mpv) and games (reshade), and it certainly didn't require AI to work.
Now in comes this company and introduces such a basic feature, while boasting about it as if it's the second coming of Einstein or something. And it uses AI, because of course it does, it's 2022 after all.
LOL.
I played with a model in a similar vein for deinterlacing old TV and DVDs several months ago. The results were mixed. Some things came out looking fantastic but some things came out looking worse than traditional methods like yadiff. I need an AI that can tell me which method is going to work better so I don’t have to do both and manually make a subjective decision.
Yet everyone assumes that Mac users dream of using their apps on some other OS. Why? I have no idea.
It's exactly the same with Pixelmator Pro. You click on the "Tech Specs" menu item, scroll to (or search for) the "System Requirements" section, and get:
"Pixelmator Pro requires macOS 11 Big Sur or later and is fully optimized for Macs with Apple silicon."
Banding is more typically seen in monochromatic gradients - black and white or shades of sky blue. But this image is a yellow/blue gradient, and it's super unnatural to have the RGB channels transition at the same time - typically they transition at different locations and that helps smooth things out.
The traditional way to do this with old Photoshop workflows was to accept the fact that if you were working with low contrast gradients over large spaces, and you were getting banding, it was probably because you were operating in a lower-bit bit depth.
So, what you'd do is simply move up bit depth for larger color ranges. There's no "machine learning" required.
What I'm getting at is, is this marketing wank? Because it sounds like marketing wank, unless they can do it without increasing file size, and staying in the same bit depth, by using some perceptually lossless dithering technique. Thus, handwavy "machine learning."
Edit: color space/bit depth
I don’t think so. There are lots of ways to end up with banding even if you try to keep it high bit depth. Photoshop quietly dithers (usually invisibly on a monitor) when going from 16 or 32 bits per channel to 8. Last time I checked, Image Magick didn’t have that option, nor do other non-Photoshop options. I feel like I vaguely recall Image Magick doing blur in 8 bits, even when working in 16 bits? Cameras can sometimes spit out banded images as well. Certainly if you didn’t control the image, someone else might have added the banding. There’s a lot of 8-bit imagery out there, and no reason to assume that everyone in the world is working in 16 bits and is banding aware.
On top of that, removing banding is obviously going to be difficult to do well, since it has to differentiate between which edges to blur and which to leave alone. This will definitely be useful and save time for working design/photography professionals.
Maybe use a zoomed example on mobile?
You can also do it on iOS by long pressing the foreground subject of an image in the photos app.
Of course it's not "magic", obviously. But it impressive technology, and it does delight users. It's taking something that was difficult, time-consuming, and complicated and making it painless.
Yes, it's marketing. The point is to get people's attention and sell the product... or at least get them interested enough to read a white-paper for more details. Marketing departments have committed much more egregious sins over the years. The use of "magical" as an adjective isn't even on the radar.