Topaz Video AI v3.1.5

here: clip
everyone is welcome to try. settings I used:
Proteus enhancement, no stabilization, no motion deblur, H265 Main10 (Intel) encoder, mp4 container

I found that when converting a DVD 25fps to 4k 60fps (Chronos Fast + Proteus), with the Replace Duplicate Frames function turned on, the final video has small jumps, jerks, and distortions that are unpleasant to the eye. Without this function, the video seems to be smoother


This is a matter of course. The function of replacing duplicate frames is to insert more frames, which of course is more prone to problems.

No issue by using H264 High (Nvidia) Here, will try H265 Main10 (intel)

No problem using H265 Main10 (Intel) as well here.

I notice thats not the clip that you posted originally.

1 Like

It seems to me that the ratio: size-quality-ease of processing, it is best to put H265 Main10 (NVIDIA) MKV on the input. Is it so?

You need to compare that in a video player. I use Media Player Classic.
Note that Video AI will always display the color wrong both left and right picture.
That is why you don’t notice the issue.

2 Likes

it is a different clip but it is affected in the same way.
the other clip I prefer not to share for personal reasons.

Have you tried other non-intel encoders?

I am new to the program. Is there a detailed description of the functions of different AI models somewhere? I try the error method, and I have to constantly rewrite the video. Efficiency is extremely low. For two months of work, I have only a couple of clips ready. In other cases, errors constantly appear in different places, due to the incorrect use of models.

[image]

2 Likes

yes, I have the issue here in MPC-BE, need to do some test, but i think it’s maybe related to the Mjpeg format you have → intel Encoder in H265 Main10, i’ll test other combinaison, hold on , and will re encode your original video too.

MVI_2232.AVI_snapshot_00.00_2023.02.19_13.34.07

You could try decreasing the sensitivity value (for the duplicate frame detection) to see if that helps (in case it’s skipping frames that aren’t really duplicate ones).

eg. the default value seems to be 10, and that might think some frames which are very similar but not identical are the same. It probably depends how much compression (or maybe picture noise) is in the frame. If there’s compression artefacts that differ between frames but the frames are really the same then setting sensitivity too low won’t be able to detect that they’re the same.

1 Like

Hold the mouse on the selected model, and the pop-up will tell you the general purpose of the model. If you’re concerned about interpolation models, you can take a look at this: Discussion about Frame Interpolation

1 Like

same with nvidia H264 High, did you tried by re encoding your video in an another format as source ? and using an another encoder as well ?

Do you think that two sentences can describe the complex work of an AI model? I read these abstracts, watched videos on Youtube. Characteristic models for slowdown, stabilization, or for computer graphics - everything is clear with this. And with the rest - complete chaos! Apollo? Chronos? Proteus? In which cases? When? Why? Nothing is clear at all. Developers should take the trouble to describe in detail the operation of each model in order to achieve maximum results.

[image]

2 Likes

same issue with using an ffmpeg Gui re encoder, 'im trying with Vegas pro 19.

If you want to emphasize the complexity of the Ai model, then I can tell you clearly that no one can explain to you what a model looks like, not even the creator, it is itself a black box. I mentioned the problems of the two models in my post, you can choose according to your own situation. In general, as the software introduces, use Chronos for small changes, and use Chronos Fast for larger changes. Apollo is not recommended.

1 Like

Found your issue !! it’s the Mjpeg format, maybe an issue with the windows decoder of the mjpeg one, i bought some years ago a licence of a Mjpeg specific decoder because of this, i had a Movie (done by a director friend) to upscale from an Avi Mjpeg Capture Card (from 1999
). Had to buy a special encoder for that, windows one was doing a mess. (it was even worst as entrelaced).

here is the picture directly exported by vegas pro 19 as Png. the issue is the same than the picture displayed in MPC (first is your original Avi source video)

(trying to play with vegaspro encoding/deconding to see if it’s related , hold on).

MVI_2232.AVI_snapshot_00.00_2023.02.19_13.34.07

MVI_2232.AVI_VegasPro19

I don’t know how you do it, but on my side, using nv encoders, there is no such problem.