Apollo vs. Chronos

Apollo or Chronos for Slow Mo?
What is the general rule of thumb (generic guide line) when to use what model for Slow Motion?
what use case or type of movie?

I know you might say, try both and see what works best for you, but that is not my question.

1 Like

Chronos because it is much faster, unless it produces obvious artifacts in which case I try Apollo in 3.0.x, which are all I have access to in v 3. Others have said Apollo produces less sharp output on some videos so really there’s no getting away from “try both and see” as everyone’s video types are different and speed matters more to some than others.

1 Like

so in other words you saying, SlowMo with Chronos, only if you see issues pick Apollo and try…correct?

I’m saying that’s what I do, because it works for the kind of old videos I am using. The bonus is that it’s faster. But I don’t ‘pixel peep’ (for want of a better phrase) like some do so I cannot tell you which is inherently of higher quality, they are both good enough for me on most videos.

Apollo analyses more frames (therefore slower) when interpolating a single frame. Apollo is better for video with more non-linear (less predictable) motion, whereas Chronus is faster but only gives best results with linear motion video.

4 Likes

I use TVAI primarily for boxing videos and have found Chronos Fast produces smooth motion as opposed to Apollo which has a slight jittery effect.

But in fact, Apollo will make the video quality lower, and produce distortion in some scenes, if you care about it, never use Apollo until it is improved.

The best approach is to take a small extract from the full length video and see what each model does with it. The models really are source dependent and settings that work on footage A may well not be as good on footage B.
All you really can and should do is try things & see what works best for that footage - the long & short of it is that there are no hard & fast rules.

2 Likes

yeah, i did that already, the thing is i didn’t see any difference on the output file(s) trying both. hence why I asked what is the rule of thumb/guide lines on when to use what module/filter for what video type/use case as a best practice.

1 Like

I use Chronos fast- manual, as it renders smooth motion in fast action sports videos. Whereas I find Apollo results slightly jittery.

1 Like

Until now, the advice was not to use Apollo at any time to avoid problems.

1 Like

that was true for Apollo v4, but since then Topaz released v5 which they claimed fixed may problems previously seen. so i am not sure this statement is correct anymore.

you should read this: Topaz Video AI v3.1.4 - #16 by q91632

2 Likes

Been meaning to do this for awhile. Here are six of the same clip done with all the combinations I can think of with the interpolation models:
Interpolation.zip link
Chronos Fast, output FPS 59.94, ChronosFast.mkv
Chronos Fast, output FPS original, slow motion 2.5x, ChronosFast25.mkv
Chronos, output FPS 59.94, ChronosSlomo.mkv
Chronos, output FPS original, slow motion 2.5x, ChronosSlomo25.mkv
Apollo, output FPS 59.94, Apollo.mkv
Apollo, output FPS original, slow motion 2.5x, Apollo25.mkv

All clips were output to png then converted back into H.265. So I cannot say the outcome will be the same if outputting straight to a video in TVAI.

I chose this clip because of how bad it is in Chronos Fast, but I’m sure it does not expose all shortcomings. Just a really common one I have seen often.

5 Likes

Chronos slomo was pretty decent. I have had the same problem as in ‘Chronos Fast’, (first section). Good to see it can be fixed with Chronos Slomo!

1 Like

On the last conversion we did, using both Apollo & Chronos (slow) to 0.96% speed on the original 25fps spurce file and then importing the output into After Effects & interpreting the footage as 24fps to stabilize it (as the output from both models results in a variable frame rate file according to MediaInfo) which then increases the run time by the 4% we lost in the slo-mo we found that the Chronos conversion resulted in a much more vibrant image than the Apollo one.
It feels like a bit of a faff to do the FRC this way, but it has proved to be the only method that results in a stable 24fps result suitable for use on an international Blu-ray.

If anyone has a better solution I would love to hear about it

Wanted to try out Apollo Fast. Did some new tests. Probably not the best tests, but here’s a link to the results. I included the source this time.
All results were passed through Artemis High Quality to remove noise before interpolation. They were set to slomo 2.5X and ‘Replace Duplicate Frames’ was turned off. (I ran them all in the CLI with PNG output and put how long just interpolation took as the end of the file name. The script I used also converted them to H.265 at CRF 22 and added the audio back in.)

When I first ran them, I had ‘Replace Duplicate Frames’ on and at the default value. It made the motion ultra choppy in some places on all models except Apollo. Turning it off seemed to fix that. I tried a few runs in the GUI not using slomo, just setting the output frame rate to 59.94fps. I didn’t notice any real differences and decided not to continue—since that’s not how I will be using TVAI interpolation most of the time.

Is “GN Original.mkv” before Arthemis or after?
Also I noticed that all the others (chr/chf/apo & apf) the picture is sharper than “GN Original.mkv”
what did you do during the frame interpolation that sharpened it?
BTW - none of them are slow motion…

Care sharing the scripts you used for those files? I want to give it a go.

Good question. It’s before.

Must be from Artemis HQ. GN Original.mkv is converted to H.265 at CRF 14. So it’s also not totally original quality.

Because I output to PNG then re-encode them to videos using the higher frame rate.

It’s a monster. As in: it’s big and messy. I have a few plans to improve it and make it more useful to people other than me. Here’s the last version with some instructions. You’ll probably have questions.
You’ll want to add all the interpolation models. In the script, search for model = 'apo-8'
Change the lines below it to:

    if filepath.get('-chr', null) != null:
        model = 'chr-2'
    if filepath.get('-chf', null) != null:
        model = 'chf-3'
    if filepath.get('-apf', null) != null:
        model = 'apf-1'
1 Like

And here’s the same clip I ran before with all the latest models and source. (Source was transcoded from the original at CRF 14. No sound this time.)
I encoded them at CRF 22, but have found that the grain stays if I lower it. I don’t like grain and need to save space on my drive, so 22 it is.

This time around I just ran the interpolation models at slomo 2.5X outputting to PNG, then encoded them to H.265 at a frame rate of 59.94 FPS. No Artemis involved.

Again, I’d say the main one to pay attention to is Chronos Fast. It’s so bad.

1 Like