Why is it that sometimes when I try to export a video, my entire PC gets overloaded to the point where my mouse won’t even move? This happens during the “loading model” portion of the export process. I haven’t done much testing, so I’m not sure if it only affects some videos that I try to export, or if it only involves Starlight Precise. (Starlight Precise is the only model I ever use.)
It doesn’t seem to have anything to do with the length of the video, as I’ve run into this issue with exports that involve less than a minute of footage.
For me, SLP often oversharpens. What helps is using Iris MQ 1x afterward—set it to “Dynamic” and increase “Fix Compression” until the image gets smoother. This can sometimes work wonders on SLP
I’m fully aware that PC users make up the lion’s share of Topaz Video owners, but if Topaz is going to continually use the argument that these models require Nvidia hardware because of their beefier hardware specs (thereby making Mac users wait longer for versions that support our hardware), then do right and either 1) give us discounted licenses or 2) give us free or discounted cloud access to these models.
I’ve been experimenting with various segments of FHD source material, and I’ve come to feel that the ideal settings can vary significantly depending on the specific source or even individual scenes. Because of this, I believe it’s difficult to rely on a single “universal” preset.
The advantage of Iris MQ’s efficiency
One of the main reasons I use Iris MQ as a second pass after SLP is its processing speed. To find the right balance for an entire video, it’s necessary to test several variations across different cuts. The light rendering load of Iris MQ makes this iterative testing feasible, allowing for a better overall result.
Adjusting for scene-specific characteristics
A setting that works for a wide shot may feel a bit heavy in a bright close-up. For example, in scenes with strong lighting where SLP might over-restore textures, I find it helpful to manually “soften” the output:
Recover Detail: Lowering this (sometimes to 10 or 5) helps suppress any excessive AI-generated artifacts.
Reduce Noise: Increasing this (around 20 or more) can be very effective at smoothing out harsh or “stony” skin textures that occasionally appear in bright areas.
Aiming for a refined balance
I treat SLP as the source of high-density detail and use Iris MQ as a tool to refine and “clean up” that information. Instead of just adding sharpness, focusing on how to subtly subtract the over-processed elements seems to lead to a more natural and consistent look throughout the video.
I hope these observations on the importance of scene-by-scene previewing are helpful to those looking to fine-tune their workflow!
Although I understand where you’re coming from, that’s not really feasible. Those of us with less powerful Nvidia GPUs are also unable to use many of the best models - should we get discounted/free software as well? You’d end up with a situation where only a minority of users with high-end systems would be paying. It would also be open to abuse - anyone could claim anything about their system specs if they fancied some money off.
Just a suggestion, but it would be helpful if I could either see the full file name of the video I used when exporting. Sometimes my video file names are along and at the end I may put 480p, 720p, etc, because I’m testing different things, but I can’t tell which video it is in the export preview. It would be helpful if you kept the mouse of the file name for a few seconds then had it show the full name.
Are you guys working on the memory leak issue that’s causeing topaz to max out some users ram to the point of slowing or freezing and or crashing when using starlight precise 2.5. I mentioned this a day after update was released but never got an answer.
Iris LQ with „soft“ settings both diminishes the somewhat „harsh“/overly sharp look and even partly jagginess that SLP creates but still keeps most of the amazing details.
Plus it effectively removes the noise that SLP leaves and by that also alleviates the „dirty faces“.
And it even makes the partly jittery motion a bit less obvious.
I’ve got even better results by first running IrisLQ V1 2x and then another 2x SLP pass. But this takes forever - so most of the time I guess I’ll settle for a SLP „minimum „ run followed by the Iris LQ 2x.
I’ve been struggling with some close-up shots under harsh, direct sunlight where the excessive details are quite difficult to suppress. I’m going to try the “Dehalo: 1” setting you suggested right away. I’ll report back with the results!
I’ve thoroughly tested that if you (taking into account the crazy render times) use SLP 4X upscale even though the output size might already be maxed out, it tends to restore text, numbers, UI on screen the best out of just doing 1X, 2X, 3X etc. (3X is the balance I guess)
The processing time is damn near 24hrs though.. lol
Check your RAM usage when this happens. (Not VRAM) there’s a bug in this version that uses ALL the system RAM sometimes. On my rig, it’s happening with Starlight HQ, but I have seen others report the same bug with SLP. (And SLP is fine on my rig)
It will probably get a patch at some point to fix it.
