Several options for you because the 4090 is so fast, it doesn’t hurt a bit!
Note that the brickwork did not benefit at all as compared to Super focus.
1 Like
Yes. I see what you mean.
I rarely find the motion correction model works on my images with movement. That’s why I was curious. It seems to add a shift to the starting image that ends up looking like a halo a lot of the times I’ve tried it.
Thx!
1 Like
The added Face Recovery seems to soften areas of the beard that Superfocus had sharpened. At least from the snip’s appearance.
Toggling back and forth I did see that, as well as his crow’s feet. I was leaning toward Refocus only as the best option.
1 Like
I am very interested in speed comparisons with your M1 Mac and M4 Mac. I hope you will do some post them here. Thank you.
I will.
I just ran a test on a 31.6MB ARW file, using the beta in standalone and with subject selection set to sharpen.
Results on the M1:
Analyse: 7"
Export: 6"
SuperFocus preview rendering: 1’26"
SuperFocus calculating ETA: 3’20"
SuperFocus rendering: 74’ (however it took maybe less than half that time)
The main issue for me with the M1 has been that 16GB RAM is now not enough to run TPAI smoothly and there are crashes etc., especially if my main editing software (ON1PR) is also running. The M4 I have ordered has 25GB: maybe I should have shelled out more for the Pro, we’ll see.
All being well, I take delivery tomorrow, and will run the same image again in the next few days and report back.
By the way, SuperFocus is way too aggressive still, especially on animal fur. Maybe this image wasn’t soft enough to merit using it:
Thanks for your report, and I look forward to the next one!
My thoughts right now:
Is the problem with Macs or with the software? ie, is the Apple chip not suited for AI-type of work, or is the software not nearly optimized enough?
My experience here with Macs is even an M2 Studio MAX with gobs of RAM is inefficient running Redefine in GPAI. TPAI Refocus can be accomplished on an M2 Mini in reasonable time, as Topaz optimized it by 2X recently, so this gives me hope for Redefine.
As I’ve mentioned previously I also have access to i9 PCs with 128GB RAM running NVIDIA 4090s and even with all that, there is still some waiting for Redefine results – though it is not anywhere near as bad as on a Mac (where it is not practical to proceed). I did some Refocusing the other day on one of these PCs and it was “not painful”.
So a PC with the above specs is my “baseline” in terms of what is needed for acceptable performance as things stand right now, but that is not entirely practical for most users, including myself (I do not own one of these!).
I have not attempted to A/B Topaz apps against others in terms of AI rendering speed, and they are all different and some process in the cloud anyway (Adobe, Skylum, etc.). FWIW I did notice that low-end, local DiffusionBee renders happened about twice as fast on a Mac Studio as opposed to a Mac Mini. But that alone would not drive me to pay double/triple the cost of a computer. 
My other issue is that my internal SSD is almost full, even though I store all my photos on external drives. This is known to slow down performance considerably. Upgrading Apple internal SSDs is a rip-off, but at least my M4 will have twice the space of the current one, which is just 256GB, with only about 7GB free! A major culprit here is On1PR’s Application Support folder which is currently a massive 50GB, even though not all my photos are catalogued. Users have been urging On1 to make it possible to shift this folder to an external drive, but so far without success.
Few people, either in the PC or Mac world, have reported rip-roaring speeds yet with SuperFocus. It is still in beta of course, and I suspect Topaz might offer a cloud-based option for that in due course.
OK. yes, the SSD can be a factor. And the 256 was known to be slower than the 500.
My experience with Super focus posted above was not bad; it’s the most usable of the new features on both Mac and PC.
I am once again getting the “invalid input: unspecified Metadata not correctly assigned” error when I try to open Photos images in Photo AI. You need to fix this because it makes your software unusable for me!
Regarding another AI task, I tried erasing the tiger using Remove Object. I got a warning about it being a large object but went ahead anyway. I gave up after just over 4 minutes.
ON1PR’s new generative erase tool (local) completed the task in just under 30 seconds. The result wasn’t very good, but using Stability AI at a small cost it would have been faster and a better.
So there are definitely some AI speed issues with TPAI on a Mac.
2 Likes
@din.begovic - I added your info to the development team ticket working on supporting the new way Apple encodes metadata in the new iOS 18. This creates issues in multiple programs. We will reach our here once a fix resolved and I will tag you here. The fix will also be mentioned in the releases notes, that we would support that new HEIC metadata encoding when we do support it 
In the meantime, you can revert iOS, or convert the files to TIFF or JPEG, and/or shoot in JPEG to forego the conversion!
would this be the tiger from your image example? If so, that is too large to remove as it’s the main part of your image. From our Docs:
Size & Distance
Keep mask sizes below ¼ of the total image size. Process each object individually when dealing with multiple objects requiring separate masks. Ensure that the maximum distance between masked objects does not exceed 2000 pixels.
It’s now not working at all and the image sticks on “locking edits” . This is on the beta.
I’ll be up and running on an M4 Mac within 24 hours and will try again once TPAi has been installed there.
Yes, let me know!
Seems to be OK now on my new M4 machine
1 Like
Quick update, now that my M4 Mac mini with 24GB of Ram is up and running.
The test image I ran before was perhaps one second faster in both analysing and exporting. I suspect the only way to find out how significant a speed increase there really is would be to run a batch and that’s not something I have the time or inclination to do at present.
SuperFocus saw a significant improvement. It took just 6 seconds approx to render a preview (compared with 1’26" on the M1), 1’28" to calculate ETA (3’20"on the M1) followed by a 34’ timeline to process, which actually came out at less than 26’ (74’ estimate on the M1).
1 Like
Thanks for the update! That’s good news.
Perhaps the M4 is getting closer to rivaling what NVIDIA is offering? I read somewhere that some upcoming Apple chip will eventually meet and exceed the 4090.
If you want to provide me with a full-quality test image of yours I can run it on the PC and get some times to compare.
Nvidia is expected to release their RTX 5000 series next month. Then the RTX 4090 will be old news. 
2 Likes
But more affordable if a 5090 is on the horizon.
