version 1.1

[QuickscopingFTW]

the transition between the windows does seem to be a fair bit better and smoother with the new 1.1 update, i see less overall shift in brightness between the windows so well done on that.
i was under the impression that the upscale step you mentioned in another post would be released with 1.1 too, did i misunderstand or has that feature been pushed back for a later release? whats the next thing you are doing?
thanks in advance

[QuickscopingFTW]

here is a comparison i did with 1.0 vs 1.1

1.0
https://github.com/user-attachments/assets/6bca6c78-0a9b-48d6-b2df-e80a2366de49

1.1
https://github.com/user-attachments/assets/05c92fe7-7e43-4a1a-a3d9-f94e694490d8

you can see that there is a lot less depth drift/shift between windows like at 20 seconds, 24 seconds especially and some others, you also seem to have improved the depth range fixing the problem of stuff in the background just showing up as black, this is obvious in the better detail of the background at 27 seconds and the better depth is also obvious at 21 seconds, it has a more metric depth effect which is good for consistency

[soyouthinkyoucantell]

Hi @QuickscopingFTW ,

Thanks for your endorsement! We are happy to see positive feedback and improved results! For the upscaling, or'd rather say, super resolution module, it was initially included together with the release of DVDv1.1. However, we decided to put it off until we train on the LTX series.

Best,
Faye

[QuickscopingFTW]

thank you for the clarification, best of luck trying to train on the LTX branch!

[QuickscopingFTW]

also 1 last thing if you are training on LTX, please can you check particles and smoke are working well/have enough training data because that is probably the weakest point of the current model, it doesn't really detect them at all currently which is a pain point for a lot of us currently, here is an example:

Example Source Video:
https://github.com/user-attachments/assets/3e4e631a-9fec-457b-972c-417ef9b651c7

Example Depth Map:
https://github.com/user-attachments/assets/780430f8-ff62-48c4-9d83-f88cfe5a212e

As you can see the particles aren't picked up at all, if you got this working well it would be a killer feature and get rid of its current weak spot.

Thanks in advance.

[QuickscopingFTW]

personally i am also testing ltx2.3 as a way to inpaint videos using videos warped with DVDs depthmaps in comfyui with some mixed but interesting results

[QuickscopingFTW]

here is another example of a depthmap made by depthcrafter which is quite good at particles vs dvd

depthcrafter:
https://github.com/user-attachments/assets/3a321c5e-ad83-45b2-9c3e-445159c69d30

dvd:
https://github.com/user-attachments/assets/263d7cd0-3f2e-4f5d-9f33-c3659a1f5870

[QuickscopingFTW]

just want to check to verify that you have seen the above posts

[QuickscopingFTW]

also i wanted to ask about the window size with ltx vs wan, while wan was somewhat limited to 81 frames i have been able to push to 500 frames using ltx in comfyui without much problem with a rtx 4080 16Gb vram and 64Gb of system ram, are you able to train it to have a large window size like that, it would MASSIVELY help with stability as foreground objects coming into view during scenes would be accounted for more and you wouldnt get depth shift nearly as much