ControlNet Deep Dive - OpenPose 1.1: Add Faces and Hands to OpenPose - Detection, Settings, and More
ฝัง
- เผยแพร่เมื่อ 27 ก.ค. 2024
- This video is a comprehensive tutorial for OpenPose in ControlNet 1.1. If you are new to OpenPose, you might want to start with my video for OpenPose 1.0, since this video focuses on the additional face and hand functionality in OpenPose 1.1.
OpenPose 1.0 Tutorial: • ControlNet Deep Dive -...
First, we discuss the five pre-processors for OpenPose as well as what the maps correspond to in terms of anatomy. Next, I show how large faces and hand need to be to get reliably detected and other quirks to detection.
Later, I cover Weight and Control Steps and how they impact your output image. Previously, the single Guidance parameter was used which has been broken down into Starting Control Step and Ending Control step. Charts for the ranges of these parameters that keep face and hand details is provided.
Finally, I put everything together to discuss where OpenPose 1.1 is good and where it still struggles.
Chapters:
Intro: 00:00
Preprocessors: 00:35
Points for OpenPose Face: 01:25
Detection for Faces: 02:22
Points for OpenPose Hands: 05:29
Detection for Hands: 06:10
Weight and Control Steps: 07:46
Ending Control Step: 08:42
Starting Control Step: 09:49
Weight: 10:38
High Resolution Images: 12:35
Review and Discussion: 13:09
![BOWKYLION - เจ้านายคะ (work-life imbalance) [Official MV]](http://i.ytimg.com/vi/6bFj8G2pOi4/mqdefault.jpg)
I felt like watching those article videos that they present their projects. Nicely detailed.
The way you breakdown these videos is clinical. Amazing bro, keep going
a video on the workflow of getting perfect hands and anatomy would be great , like mixing depth maps , open pose and best ways to get those maps ( blender rig , photoshop etc) , Very helpful content , almost no one is covering this , can you link if there is official description about different preprocessors
I don't think I could make a video on perfect hands and anatomy in good faith because afaik, especially with hands, it's still a numbers game.
Tutorial on multi-controlnet sounds promising. I might look into that.
If anywhere has official desciptions of the preprocessors it would be from ControlNet's creator, but I don't think there is.
github.com/lllyasviel/ControlNet
There were two things I was really hoping for in 1.1. Reliable hand reproduction (which as you have noted, has failed). And second, reliable reproduce of facial expression/emotion (happy, angry, shouting, rolling eyes etc) I was not able to either. Can you give some tips on whether you are able to?
For facial expressions, you can get some details from the map that help, (mouth position, etc), but prompting is still needed. I've found that active descriptions (yelling, clenched teeth, frowning, etc) are better than passive descriptions (angry, sad, etc).
Eye position works sometimes, but I think there are detection issues Some images can do it consistently at high weights, but others with very similar eye positions just don't work
Tfw your videos are so detailed and well-tested that they take forever to get produced and released 😞
This is the way
@@siliconthaumaturgy7593 This is the way
It's a step in the right direction but has a long way to go. Still far too many deformations in the result. Faces I'm not fussed about but hands and limbs are a mixed bag.
Also the editor needs some work especially when you're trying to adjust the skeleton.
That said I watched a video on DW_Pose preprocessor that generates far more accurate results, may be worth looking into.
I still haven't found a good editor or way to generate maps. The few good riggers I've tried sadly don't output to OpenPose.
are these detections in 3d space? any way to adjust hands manually if detected poorly?
Is it possible to use existing skeleton poses with a different image for the face? Do I need to manually merge key points with skeleton?
I haven't experimented with that, but it should be possible.
However, the basic skeleton has points for eye and ear position, so you would probably need to change or remove those to prevent conflict with the new face.
Bro can you make a video about merging models, mainly i want to know about MBW and what each sliders do when merging models.
I was using supermerger and model block weight extension and tried to merge models but due to lowvram i couldn't finish my experiment because it was taking so long to merge and then generate images.😅
I appreciate the suggestion, though honestly I haven't messed around with merging models much so I'm not super confident in my ability to deliver quality content on it.
I still dont know how to use open pose
>0:33 🧐
>CIA?