Selur's Little Message Board

Good day!
Tell me, is it possible to somehow increase the smoothness of camera movement (panning) of a video, but at the same time not change the movement within the frame itself, that is, duplicate the frame and only draw the edges when moving?

Not sure, I really get what you want to do. Don't really know what you have and what you are aiming for.
Do you have a panning shot or do you want to create one? (Ken Burns? Do you want to create something like https://forum.doom9.org/showthread.php?p...ost1954945 ?)
If you also need something like content aware filling (outpainting), you will probably not get far.

You probably will have to write a script for yourself or use another tool than Hybrid.

Cu Selur

I have an anime, when the camera moves the picture strobes. It is very noticeable. 
I would like to increase the smoothness of the camera movement. But so that the content of the frame does not change.

---

Conventionally speaking, get from two frames

Three frames, with different camera positions, but with the same content

So that there are no attempts to finish the frame itself, like here for example

Sounds like https://github.com/routineLife1/DRBA might be what you need.
For Vapoursynth this can be used through: https://github.com/TensoRaws/ccvfi

Cu Selur

Thanks for your reply, I'll try it!

As I understand ccvfi can be connected via VapourSynbth inside Hybrid? 
Or am I wrong. If it is possible, can you tell me how to do it?

Hybrid has no option direct support or add-on for ccvfi.
One could install the torch add-on, then install ccvfi and then could use it through a custom section. (I did that before to do some testing. If I find some time after work, I'll write how I used it.)

Cu Selur

Had a few minutes before I'm off to work.
Here's how I used it:

• Installed Hybrid with torch-addon as a base.
  
• Opened a Windows Command Prompt inside Hybrid/64bit/Vapoursynth
  
• Installed ccfvi through:
  
  Code:

  python -m pip install ccvfi
• Started Hybrid, opened a new custom section (somewhere below deinterlacing) and used the following code:
  
  Code:

  # requires colorformat RGBH

from ccvfi import AutoModel, BaseModelInterface, ConfigType

model: BaseModelInterface = AutoModel.from_pretrained(
    pretrained_model_name=ConfigType.DRBA_IFNet,
)
clip = model.inference_video(clip, tar_fps=60)

# framerate 60
  (first line is to let Hybrid know that this section requires RGBH, last line it to let Hybrid know that this section changed the frame rate to 60fps)
  
• normally use Hybrid.
  

Cu Selur

Ps.: Not sure whether I'll add this to Hybrid, since I'm not sure how common this problem is and whether it is worth the effort to add it. I just checked whether it worked, didn't do any comparisons or similar whether it was better than using RIFE with different models or similar.

I tried to do as you said, but something doesn't work.
Torch Addon is Torch_VapoursynthR70_2024.12.27 from Google Drive, right?
It needs to be copied to %Hybrid%/64bit/Vapoursynth, right?
Then I do as in the video, tell me what exactly I'm doing wrong?
https://youtu.be/qg_CFylkY4o

Quote:It needs to be copied to %Hybrid%/64bit/Vapoursynth, right?

No. The torch-addon contains a Vapoursynth folder which should replace the original Vapoursynth folder.
Thus, you should delete %Hybrid%/64bit/Vapoursynth and extract the torch-addon into %Hybrid%/64bit/.

Quote:Torch Addon is Torch_VapoursynthR70_2024.12.27 from Google Drive, right?

Yes, assuming you use the latest public version and not the latest dev. If you use the latest dev, you also need to use the torch addon from the 'experimental' folder.

Looking at the video:
1. you should reset the FilterOrder, save the defaults and restart Hybrid like the log mentions.
2. Okay, problem is that ccvfi wants to download the models. (totally forgot about that, since I use Hybrid as a portable version)

To fix this, you can download the models manually and place them under %Hybrid%/64bit/Vapoursynth/Lib/site-packages/ccvfi/cache_models/.
(models can be found in the 'model zoo' on the release page)

Cu Selur