1000 Boudoir Glam shot Poses
V1 Used the new "BODY25" 25 point Skeleton extraction methods which is a more advanced and accurate method, but the skeletons used in openpose for Automatic1111 is based of "COCO" 18 point skeletons, I didn't know this so I re-extracted everything. My extractions works better now.
I was experimenting with Openpose and extracted myself a few Poses from a pdf I found on Boudoir and Glam shots, thought it turned out not too bad so I'm sharing it with you guys.
I saw some poses with included hands as depth, if anyone has some feedback how to get that working I will definitely update this collection.
I have more reference images that I will add as soon as I have processed the datasets. This dataset's original images can be found here:
Description
V1 Used the new "BODY25" 25 point Skeleton extraction methods which is a more advanced and accurate method, but the skeletons used in openpose for Automatic1111 is based of "COCO" 18 point skeletons, I didn't know this so I re-extracted everything. My extractions works better now.
FAQ
Comments (15)
Thanks for the update.
I wish there were more example images, its sometimes hard to describe the pose in prompts if you aren't sure what it is.
Tacot, in the details of this page, i left a link where you can download the image pack i ised to extract from bud :)
insanely helpful. Thanks!
Is the format of the coordinates in the JSON files compatible with Automatic1111? I downloaded version 2, but it looks like the format of the coordinates are completely different. I think the open pose editor is expecting sets of coordinates in pairs (X/Y). The files in version 2 have something completely different.
How do I add it to automatic1111?
happy to get a two persons version, thx for uploading your obviously ^^ work
I am new to this, I have learned something, but I would like to know how to use this in automatic1111
I want to know it too <3 hope someone came and help us
to use these in automatic1111 you need the controlnet extension loaded. I am not sure if you still need to manually download files anymore, but once it's installed you can click where it says control net - under cfg scale.
Click enable, click where it says drag or upload, upload a pose you like from this unzipped collection. Then drop down "preprocessor" and pick openpose full or openpose. The model should say openpose - mine says control_openpose.
You can mess with the other settings but you probably don't need to. just make sure that your txt2img or img2img resolution kinda matches the pose image resolution, otherwise stuff gets real weird.
So instead of using controlnet to extract a pose from an image, you use it to extract a pose from a pose, it's very fast. The first time you generate using controlnet with each model it will download stuff, 200-600MB each time.
I don't understand. Are you saying that we're supposed to input the image of a pose as an input image that will be run through with ControlNet OpenPose preprocessor? I tried that, and it outputted a black image that contained no pose.
what are the .json files for?
.json files are for general info that you write in Automatic1111
The program will generate them anyway. So I think that there is just some info like descriptions. But I may be wrong here.
So just open them with a notepad and see what inside XD
I was unable to get these working. There was no instructions how to use them in the "model card", and the "i" link next to "poses" category definition which had a tooltip "how to use" lead to a page which did NOT include any instruction about how to use the poses. I tried to use these in A1111 webui ControlNet extension with the OpenPose option, but it did not work.