idea from this reddit: https://www.reddit.com/r/StableDiffusion/comments/1i4fkuk/hunyuan_vid2vid/
custom node:
lora used in example:
https://civarchive.com/models/1120059/jennifer-lawrence-hunyuan-video-lora
https://civarchive.com/models/1131159/john-wick-hunyuan-video-lora
how this works, step by step:
Locate the specific video file you want to work with.
Convert the video using FFmpeg. Use FFmpeg to change the video’s dimensions and frame rate. A recommended frame rate is 24 fps:
Update the node’s dimensions to match the video output. Ensure these values are multiples of 16.
Refine your prompt.
Specify both source (original content) and target (desired content or style).
modify the target prompt to include the elements you want changed.
Add a LoRA model to further control or style the output.
Experiment with different denoising settings. This often requires trial and error to achieve the best balance between preserving original details and introducing new stylistic elements.
Description
FAQ
Comments (13)
seems, interesting, can't run it right now, but is it a good old vid to vid (I mean a real one, meaning you plug any vid you like as input) ?
Can this follow facial expressions too somehow?
All I get is blurry results, I'm using the FP8 model, maybe that's the issue? Maybe I have to use the bf16 model
Hi I'm a complete noob at comfyui. How do I install the HYFlowEditGuider, HYFlowEditSampler, and HYReverseModelSamplingPred nodes? I don't see them in the comfyui manager.
good,but it's different as other swap face tool?
is there any way to adapt this to use teacache etc?
I just get a black video, what am I doing wrong ?
I download the workflow and it works without error report.But there is a problem,I used a Thanos lora from civitai,but no matter how I change de promte,the output video looks very similar to origin video,and it seems like that Thanos lora did not work any more
Is there a way to increase the duration of the output video?
Thanks again for this workflow it works well
i for the life of my can't get anything that's not blurry. Even with default settings. Any tips for this maybe?
the workflow is working fine. I want to try v2v with lora I trained on myself with the severance john wick video. I broke down the 20 second clip into 5 sec each
for each diff clip I'm getting different face, how do I maintain consistency over entire generation ??
all settings are default as in the workflow by default
Yeah I'm just getting the same video output as the original file even though I have a character lora referenced with trigger word... what am I missing?