v1.2 finally out.
all instructions + custom node info are included in the workflow readme section.
v1.1 is out.
Some minor improvement on face morphing/degradation issue.
Minor improvements on scene transition.
Please get: https://github.com/ltdrdata/was-node-suite-comfyui and https://github.com/wildminder/ComfyUI-KEEP for v1.1
As requested by many of you, here is my video extension workflow.
With this you can infinitely extend your videos to your liking.
Have fun and join my discord to contact me directly.
Custom Nodes:
https://github.com/pythongosssss/ComfyUI-Custom-Scripts
https://github.com/yolain/ComfyUI-Easy-Use
https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite
https://github.com/cubiq/ComfyUI_essentials
https://github.com/banodoco/steerable-motion
https://github.com/Fannovel16/ComfyUI-Frame-Interpolation
https://github.com/willmiao/ComfyUI-Lora-Manager
https://github.com/Smirnov75/ComfyUI-mxToolkit
Description
Minor improvements on face morph/degradation issue
Minor improvements on smoothness of scene transition
FAQ
Comments (28)
I seem to be having an issue when running your great workflow:
"E:\ComfyUI\custom_nodes\ComfyUI-KEEP\modules\keep_model_loader.py", line 95, in load_keep_model_pack
raise RuntimeError("KEEP architecture class not found in ARCH_REGISTRY.")
RuntimeError: KEEP architecture class not found in ARCH_REGISTRY.
Has anyone had/have this issue and a possible fix?
Cheers 😊👍
@PixelBlitterBoy its an issue with python version on the embedded version of comfyui
if you are not comfortable messy with the python version, i'd recommend disable the KEEP nodes for the time being. will work fine without
@Chriqro Many thanks for the quick reply 😊👍
@Chriqro - ComfyUI-Keep looks promising for faces. I must assume you've been able to make it work. Could you possibly share your python setup (version) and suggest a solution? I have four different venvs with different python/torch/cuda setups to play with, and I am also getting the same error.
KEEP architecture class not found in ARCH_REGISTRY.
Do we need to install the original KEEP from https://github.com/jnjaby/KEEP ?
Otherwise, fantastic work, as usual.
@arsengo i'm on python 311 and i did not use the phthon embedded version of comfyui.
Solving this issue does help a lot with character consistency. This is how I solved same error.
1. Delete the ComfyUI-KEEP folder in Customer folders.
2. In Command Prompt, go to app folder in your ComfyUI install folder.
3. Activate the venv by typing: .\env\Scripts\activate
4. Now go to Custom Nodes folder within the same command prompt window.
5. Follow installation step in the Github page. https://github.com/wildminder/ComfyUI-KEEP?tab=readme-ov-file#installation
6. Make sure to restart ComfyUI
7. In my case, I had to manually create "/keep_models/KEEP/" folders in ComfyUI/models
8. I also had to manually download the model as running pip requirement didn't download some models "detection_Resnet50_Final.pth", "KEEP-b76feb75.pth" (I'm not sure if both are needed but I have downloaded both.
@Sam110177 not sure what I'm missing, but this does not work:
3. Activate the venv by typing: .\env\Scripts\activate
all I get is The system cannot find the path specified or this in powershell
The term '.\env\Scripts\activate' is not recognized as the name of a cmdlet, function,
script file, or operable program
You're meaning the ComfyUI\app folder right?
how do i install these custom nodes from github pls help?
you can use comfyui manager if you aren't familiar with how to install.
alternatively, you can just google it or look at the how to install instruction on each custom node github page.
The results of this WF is mindblowing! OP, could you please give a short explanation what is actually happening here? Does this basically makes a new i2v from the last frame of the input video, then interpolate between them? Or is there more at play to influence the extended video?
For anyone who wants to use GGUF: 1. Add your GGUF loaders; 2. make sure to expand each loader's "Set Node" to ensure the variable names are correct after attaching the new loaders; 3. Up in the "Process" group you may need to expand some of the "LOAD" nodes to update models.
@altoiddealer yeah that is basically it.
it takes the final frame of the last video, upscale and detail the face slightly then runs a i2v generation based on your prompt.
As awesome as the results are, I am experiencing an issue where the overall colors change, including the input segment. With every successive generation the tone is becoming more and more red. At the end of the day I can just apply a filter to the final result to try and bring it back, but maybe you have an idea what causes it?
Edit I think it might have something to do with merging the videos before applying the VAE over the combined video? I'm not quite a pro yet so scratching my head on how to try "finalizing" the extended portion before merging them.
Edit2 Just seems to be an issue with WAN / VAE - or something. When generating any I2V it seems like the first frame is always more saturated than the input image, even if CFG = 1
Sorry for the comment spam - I also want to point out that the slider labeled "Low Noise Steps" is mislabeled as this value represents the number of High Noise steps.
@altoiddealer The labelling depends on what you put for the model loader.
Re the color issue, I’m aware of it and am looking into solutions.
@Chriqro From my testing, all Wan 2.2 gens have some color inconsistency in the output. When using lightning loras, the cfg must be 1.0 otherwise the issue is stronger. CFG does not seem to make the effect worse when using no lightning, though.
Thanks for this awesome workflow. Any chance you can add audio in your original Lazy and this workflow using NSFW MM Audio and Lipsync. That'll complete it.
@Sam110177 down the line i think. looking for resolve face consistency issue first before looking into audio
@Chriqro i saw a workflow that included roop to help with face consistency. I just don't know where that workflow is. Would that be a solution because I have the same problem
@puthenpurackalst6910 check out the v1.2 workflow for video extender.
face consistency issue is drastically improvement with the wan IPA nodes
This workflow is just great. Thanks.
I have this error when trying it, can someone explain how to fix it ? "Model in folder 'upscale_models' with filename '4x-UltraSharp.pth' not found."
if you have comfymanager installed, go to the model manager, sort by base: upscale. and it should be in that list. if not its probably on huggingface.
figured out thanks !
nice work. thanks for this beautifull wf
Great workflow, but i have face degradation pretty often, maybe cause i'm not able to generate hight quality vids ? Any tips to avoid it ?
Great workflow! However, the ComfyUI-Lora-Manager plugin is redundant—it isn’t as useful as the pysssss's model/lora info feature in ComfyUI-Custom-Scripts.
Why does every artist use different custom nodes? We have loading cmd longer than the bible soon