Thank You Master..

[cyberbol]

Thank You for your hard work. I really cant wait to try it, but i getting this error:

SamplerCustomAdvanced
CUDA error: invalid argument
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

EDIT:

Its working when i changed to GGUF MODEL. So probably i have something wrong with standard models

EDIT2:

Unfoirtunatelly with GGUF i recived also Oout of memory

I have rtx 4090 24gb vram and 96gb ddr5

[RandomInternetPreson]

Interesting, I've been noticing issues with the workflow myself. I traced it to an issue with ComfyUI. From your screenshots, are you hitting the "play button" while viewing that layer of the workflow? Because I found this weird glitch where I needed to be at the top layer, the window where you enter the prompt; else too I would get an oom because the prompt enhancer node would load to gpu and not cpu memory.

If that doesn't work, try the workflow from the swan video. I'm curious to know if these suggestions help fix anything.

[RandomInternetPreson]

Also, try turning on the --lowvram flag to see what that does. I think it helps solve the issue where the enhancer node is being loaded to the gpu sometimes.

[cyberbol]

Thank You for your reply . but unfortunatelly i still have OOEM using your WF and your node. Trying generate exackly same video as you on RTX 4090, 96gb RAM.

Got an OOM, unloading all loaded models.

Tried use a --lowvram, tried use your new faster node update.

[RandomInternetPreson]

Greetings! Alright, so I've been thinking about this one, with a 4090 and 96gb of cpu ram, you should be able to reproduce the workflow from the LTX-2_00015_.mp4 video.

I have a couple of thoughts:

1. I've updated the code, have you tried the newer code, or the code from the "FasterVersions" folder?
2. If your system has only 1 rtx4090 and your monitor is also running from the 4090, then you may not have enough vram to reproduce the output from the workflow exactly (your monitor will need vram to display). Even if you don't have your monitor plugged into the 4090 you still might be a tiny bit short. The reason being is that your operating system/drivers need some vram all the time. I'm on Ubuntu for example, and there is 0.486 GB used on every card all the time. If I switch to different drivers the permanent vram consumption will almost DOUBLE! I don't know what it is on windows or the different flavors of windows/linux, I just know that the system VRAM consumption can change a lot from system to system and if you truncate things down by like 50 frames you might have more success?