Hm, in my case the processed frame rate raised from 0.9 without installed to 1.3 after installation and GPU usage raised up from 50 to 80 percent. I am using Artemis model for denoise 4k video.
FFmpeg is not one who uses, but TensorFlow or PyTorch is silently using it when available (not sure which engine is used here). This lib is implementing some common functions used for AI processing optimized for tensor cores of new GPUs.
3.0.5.0.b upscaling progressive 1280x720, 50fps, AVC, and mp4 to 4k, 50fps by Proteus Auto, H265 Main10 (Nvidia), and mp4 I can’t report any difference to 3.0.4 before. All in all same behavior, same output quality. [AMD Ryzen 7 5800 32GB, RTX 3060 12GB] 0,69 spf, CPU ~12%, GPU ~40%
I like the direction VEAI is going, but to be honest I mostly use it for upscale videos. For color/contrast correction and stabilization of video clips, I use prodad Mercalli v6, which is specialized for stabilizing recordings and also has an AI mode. For video interpolation, Flowframes-Rife which in the new version has vaporsynth support and interpolates frames extremely fast. Finally, I upscale the video with Topaz VEAI.
I agree. I wish VEAI would just stick with upscaling videos. There are plenty of other tools out there that can do these add on things already, better and faster.
worth mentioning:
Selur’s developer build of Hybrid with the installed torch and vaporsynth addons and the vsgan deep learning models pack.
there is open source project “chainner” for image and video processing with many possibilities.
The biggest downside…to both== certain models like the SwinIR 4x-L requires an decent GPU…
so simple upscaling can take hours and hours,
And another big … they are not so GUI user friendly…
Topaz not have these problems.
I sincerely believe that VAI devs team will continue to train and improve its models to perfection.
I have very poor performance with a new RTX 4090 - only ~20% GPU utilization reported by GPU-Z and also no more than 6GB of VRAM being used… any ideas?
I hope someone posts some ideas. I was going to update to a 3090 from a 3070 and don’t think it will improve my speed much since you have the 4090 and not seeing high gpu utilization. Seems like the 3.0 version is slower from the posts I have read.
With 3.x version I’ve a quality issue using mp4 output. 4k max bitrate and it has blocking (compression artefacts). With 2.x version everything was fine. Anyone with the same problem ?
I don’t think he has integrated the mps pytorch acceleration for Mac into the torch addon but it’s definitely best to ask him for fresh links on his forum
Under “Enhancement” there is a slider and input field “ADD NOISE”.
You can only set integer values.
However, when using the ffmpeg command line, you can also enter fractional values (i.e. noise=alls=1.5).
The GUI needs to be fixed.
Oh goodness. That was a lot of confusing things to install. I have yet to run a test to see if it was worth it, but I can say it was not for the faint of heart.
I would like to see a models for enhancing old films. Such as removing scratches, dirt, faded colors, emulsion damage, etc. Also, how about a colorization model for black and white films?
Dragging the slider across the timeline (anywhere it’s available) takes 50s to a minute to update, even just a moving it forward/backward a single frame, frame stepping still works fine though until you step backwards to an I frame (this is on a Gen 4 NVME drive).
Oh, and installing version 3.0.4 on top of 2.6.4 removes the ability to open .avs in the older application.
i did a post on the forum about this version, however staff dont seem to be taking any nothing of it.
im rather iritated by any company that dont get the basic fundamentals on point, and instead they include other features.
the v2 had its issues which were never resolved, such as the preview period of time.
and now in this version as my posts stated, there are a NUMBER of issues, that should never be issues and the v3 shouldnt have been released until quality control was done.
ive just come away from a near 4 day render. only for it to stop because it said i didnt have enough space. and yet i had near 4TB available, and just before the end of the render i checked the file size and it had only gone above 2TB. so how on earth did it suddenly not have space?
No clue how such basic thing as “list” or “details” view mode wasn’t included. Thumbnails are ok, but info about file as well - for example i had few versions of 3D render with different FPS, no chance to know with one is selected at current state
if you look in the forum, ive pointed this issue out along with several other problems
it really is as if they are trying to appease the crowd wanting features, rather than getting the basics on point each time and improving the function and quality control everything before it goes live