Hoping to get some more love for AMD, AV1 option still not enabled for 7900 xtx unfortunately. I posted the most relevant benchmarks for the typical all AMD build. Also nice to see that the 7900 xtx perform similar to the RTX 4090 but cost significantly cheaper. Also would be great if you guys updated the software to leverage the AI accelerators in the new AMD cards similar to how the software utilizes nvidias ai cores.
Also, here is a video of it happening during a HEVC upscale. I have no idea why I have task manager opened to iGPU, but you can see what the rest of the system is doing. This takes place approx 6 minutes into the render.
Another reason why old models should be available. For some videos Iāve gotten better results with older models so it would be great to be able to download them and use them for certain tasks. Iām also wondering if some models are being overtrained ā¦
how old are you talking about? I wonder, if you install an older version if they are still there? Still use 2.6.4 because i ran out before 3.0 and my trial of 3.0 was such a disaster. Not just bugs, the UI was terrible. 2.6.4 goes back to Artemis 10 for them (up to 12 or 13). It says older versions are not available. Maybe older versions have even older models?
Gonna try 3.1.10 out again and hope itās ui treats me at least a bit like a grown up (=not forcing 10 āassistantsā down my throat every turn). Model and performance changes (i thought before this) are probably so many now, itās unavoidable to bite the UI bullet.
I confirm for the preview, it has been bugging for quite some time now, when we get to a point, it shifts the source image by one or two images. hard to compare in good conditions, I prefer the preview of 2.6.4. but heyā¦ apart from that, the latest versions are faster and I noticed especially with gaia when I upscale the video.
Artemis V8 and earlier is what your referring to. You can import V8 into version 2.3 of Topaz, but it was around then that V9 was done and broke it. You canāt import models into version 2.4, so consequently 2.3 is what I still use currently - and this was one of the reasons. The other is also because V3 has no frame control yet for exporting and side by side comparisons on preview are still borked.
You can also use any of the Artemis versions in Version 1 of Topaz as well as they all pre-date it - I have played with Artemis back into 1.5.3 of Topaz.
The main downside is that my scripts that worked well with Artemis take a long time to export. As the current version still wonāt bring in Avisynth imports, I have been trying to replicate the results in Proteus so I can export much quicker and then process in Topaz.
The export only would usually take 6x hours. In V2.3 this is fine, because as it exports it upscales, so the overall time is greatly reduced. To do the same thing in current versions meaning adding 30-40% more time to process an episode, putting aside the fact that Artemis doesnāt work well anymore.
Results are mixed still, but I have placed some examples in the star trek thread recently, which I think replicates much of what Artemis āused toā do - unfortunately its such a very specific thing where different videos may have such very different Proteus settings that copy pasting is almost useless unless your working with the exact same source file.
Hi, thanks for this Version 3.2.0.4.a - but unfortunately my MacPro 2019 still crashes.
Sending you 2 logs, mayby it is revealing: logsForSupport.tar.gz (968.6 KB) logsForSupport.tar.gz (241.7 KB)
when using video AI 3.1.10 I was trying to upscale some old tv shows that I had and notice when viewing the preview in and external video player there is about 5 seconds where the video is frozen and only audio plays from the video, but of frames before the start of the preview frame, then when the audio catches up to the moment where the video is frozen the video will then play with the audio. This would make it impossible to upscale old tv shows.
Thatās because when making previews, it cuts at key frames. Sometimes there is a lot more audio in the cut than video. When doing a full video, without cuts, there should be no such issue.
deinterlace doesnāt work that good think it was better even in 264, maybe you could look at it and add more options since there are many ways to de interlace thanks.
I highly suspect its related to the change in no longer handling frames of the image - one the things that can kill my usage of V3 still. In V2 the sequences were handled by frame number not time index, and I suspect that change is what has broken the previews and why its taking so long to fix it.
My guess is they need to fix the frame handling before the previews can be fixed, and given that the frame handling was changed entirely in V3, its probably a very large change. Having said that, someone elseās work flows may be able to enlighten me, but I have seen no advantage from changing from frames to time index - it only seems to have made everything harder/worse.
Is there an underlying reason that is more technical I donāt know of? Otherwise, it would seem to make more sense to drop the time index handling entirely and revert back to frames. If it hasnāt been fixed over a year after starting V3, it feels like it was a bad move from the start.
