I would use Proteus and fine tune it, that is what that is for. You can mess with the Reduce Noise, Sharpen, and Deblur (you will get some blur when you denoise). Pretty much get to a scene you want to work with, change enhancement to Proteus, hit the Estimate button… then fine tune from there doing a new preview after each change. You can dial in MOST high quality movies with grain pretty well removing all grain without making them look like rubber.
Ah thanks, I thought it was a fault with the program itself and that it would translate to exports.
I also use proteus with the manual sliders. Every VHS mp4 have their individual nuances so I’ve found that doing it this way at least optimises improvement. I have in the past overcooked things to the point of plastic android faces.
Thanks! Try! The Proteus Manual has not yet reached…
Hello! The has been updated for Video AI 3.2.2, and can be found here.
Yes sir. Been reported many times, going back to original version 1 release. My son jokingly referred to these mysterious aberrations as “Floating Q Spots,” because they appear in many Star Trek Voyager episodes, but limited to dark areas in the frame. They seem to jump around when played back, because they morph into different shapes with each upscaled frame.
Tied that did that but No Useless anyway if you do a preview You willl see proteus does it faster than Theia I am very sure that proteus does a count of 20 frames than does its magic Theia seems to do it frame by frame
As a senior who hires techies every day I think he might the team lead to perfection
See this interview from 2019
QuarkWorks: Meet the Team Series — Jacob Muchow | by Hannah Pratte | QuarkWorks, Inc. | Medium
Hi, i’m working on vhs acquisition and first pass is deinterlace, before to color correct and upscaling.
n 3.22 i saw a worst result, 3.1x work fine, smooth picture, last update give me on same picture a lots of artefacts, strange dectails and harsh picture on all three option of deinterlace.
huh really? never saw that before. Infact the very same video is working just fine in Version 2.6.4 but not in any of the 3.x versions i’ve tested.
myorders Bell 214B1 Driver, what preset do you run? never seen this abboration
Thank you. If possible for the future, please include the link to the script with each new TVAI update.
It has been occurring since 2.x.x.
This phenomenon occurs at 4x scale.
You really need to ask d.raebiger, because I haven’t been using TVAI since my license expired long ago. It’s been so long ago now that I have no idea what preset/model I was using at the time. My problem happened long before version 3. I’m in that group that believes TVAI is still beta, and folks really shouldn’t be charged to provide feedback for software that is far from being ready for prime time.
Why not use progressive,what model are you using？
VHS is interlaced source, i should use interlaced models or i obtain a worst lines of half fields.
i tried all three option that i think are dione models. in 3.2.1 work fine
now i keep 3.2.1 version and convert from interlaced to progressive, than later i will did more test.
Hm not for me. I have it for FHD->UHD (2x) as well. And not in Version 2.
Been using Topaz Video AI from very early on and this is my first post. My Main use is the cleanup and noise reduction of low quality and low resolution video of various formats with either lots of noise or compression with an emphasis for recovering as much original detail as possible at an individual pixel level.
One of the models I had the most impressive results was with Apollo 4 with just by doubling the frame rate without any other filters. It was able to bring out accurate details and had a noise and compression reducing quality that seems to have been lost with later Iterations.
Although many of the posts pointing out the lower quality in some of the faster models have not been helpful there does seem to be an observable and fairly significant loss of detail fidelity in the latest Apollo versions. Specifically in a closeup examination of interpolated frames the pixel values seem to be ghosting over each other causing details in motion that would normally be occluded by other details further towards the foreground to continue causing the two areas to overlay with a transparency or lower opacity causing a major loss of detail
I saw the ability to use older models was being planned for future updates and I really hope that comes soon. I get that most use cases are for upscaling and fine tuning high quality video but I hope that cleanup and detail recovery of low quality video doesn’t get left behind.
I really think Apollo 4 did something special and I think the engineers should give that model a second look as a low quality detail recovery tool
Guys I love your software, please don’t prioritize speed over quality of the result. I get there needs to be faster models for the people who need that but don’t leave behind accuracy and detail recovery. I’m a huge fan of your software. You guys are doing amazing work and can’t wait to see what comes next.
It has been occurring since ver 2.x.x and also in v3.2.2.
However, this is the first time I have heard of it occurring at 2x scale.
Please continue in the bug report thread.