image scale DLSS_5
video DLSS_5 + Interpolation + additional upscaling (optional)
Interpolation and 1x5 upscaling DLSS_5 take 44 seconds from a cold start. 5sec video. (1152 x 640)
DlAA = 20sec
Don't look at the 3080 Ti.
But my card isn't an ordinary one—it's a rare engineering sample.
It actually hits 1900+ in boost and catches up with (and sometimes even beats) the 3090ti.
as an example minimax H3
and that's without overclocking
Yes, the hardware is simultaneously rare, unique, powerful, and cool-running—turns out that’s actually possible.
Works on 2000\3000\4000\5000-series nvidia cards (windows only)
installation
1) Locate config.ini:
Navigate to your ComfyUI installation folder.
Depending on your version, look inside
ComfyUI/user/_manager/,ComfyUI/user/default/ComfyUI-Manager/, orComfyUI/custom_nodes/ComfyUI-Manager/.
Edit
config.ini:Open
config.iniwith a text editor like Notepad.
Under the [default] section, set:
security_level = weak
allow_git_url_install = true
Save the file.
2) Drag and drop that downloaded JSON workflow file into your Comfy.
3) Install the missing Nodes via the mod manager, then restart Comfy.
4) After restarting, open the mod manager, select the "Install via Git URL" option, and install the additional node package using the link. https://github.com/orex2121/ComfyUI-DLSS5-orex.git
(link to repository GitHub - orex2121/ComfyUI-DLSS5-orex · GitHub )
5) Restart Comfy again after everything is installed.
6) Activate the node pack by clicking the blue button and pressing the icon.
Conversely, block the pack with video nodes.
7) Drop any image into the image upload node and click Start.
After that, the DLSS 5 file should start downloading; you need to wait, and the upscaling will take place.
If the download fails for any reason, download the nvngx-v1 file from https://github.com/orex2121/ComfyUI-DLSS5-orex/releases/tag/nvngx-v1
and place it in ComfyUI\custom_nodes\ComfyUI-DLSS5-orex\runtime\
You can now make full use of this workflow.
Do not enable the additional upscaling option unless you have at least 48–64 GB of RAM. It yields better quality at 1440p but takes significantly longer to process and requires a powerful computer. By default, I recommend using 1080p in the video output node (if the input is 720p or similar) combined with DLSS 1.5; this is very fast and should work seamlessly for most users.
I recommend using DLAA and the additional upscaling option for 2K resolutions and higher, or simply if you are using interpolation, are satisfied with the current video image, and just want to add DLAA.
The default interpolation setting is 2x.
interpolation and upscale (first workflow)
A simple but effective upscaler and interpolator (increases FPS).
If the original video is 24 fps, set the interpolator multiplier to 2 or 3 (2 for 48 fps, 3 for 60 fps).
Set the FPS value in the Create Video module (48 and 60 fps).
For upscaling only, disable the interpolator and write its original frame rate in the video creation block.
In this example, we upscale and interpolate video from Wan.2.2 368x640 fps 24 to 736x1280 fps 60 and to 1472X2560 fps 60
Description
FAQ
Comments (3)
Interesting workflow. Simple and efficient. Except that it shortened my video (with RIFE VFI disabled). Is that normal?
No, that's not normal. I forgot to mention that without the interpolator, which increases FPS by creating intermediate frames in the video creation module, you need to set the FPS to the same as the original video, then Warflow will work purely as an upscaler. So, if you have a video at 24 FPS, it will speed up if it's recompiled to 60. Instead of disabling the interpolator, you can simply set the multiplier to 1.
I dont understand why anyone still use non Tensorrt Upscaler/Rife. Until there are some special models -there os no sense in that