Sign In

Stable Diffusion: ControlNet - From Concept to cool Image using OpenPose

Objective

Testing the ControlNet plug-in with the OpenPose plug-in from an idea to a complete image.

Preface

I will show you some of my tests about this cool plug-in, it is not a perfect guide and doesn't cover everything. This is an easy way to share my test.

Environment

For this test I will use:

Making an image

I want to make an image of one woman jumping in the desert.

Prompt: 1woman blonde wearing (white shirt and jeans:1.2) black boots. Jumping in the desert (full body:1.3)

I set the size to 768x512 and I run 4 image generations.

I run and I decided that I like this image here.

I like the image "idea", but not the face, the background and the clothes.

Now I drag (or I open the file) to the ControlNet Single Image box.

Then I set Enabled and clic on openpose.

When you click on the (1) button you can see the openpose image

Look like OpenPose didn't recognize one leg).

Now If I clic generate and can see the position of the woman.

Let's change the prompt : 1woman blonde wearing (red bikini:1.2) . Jumping in the beach (full body:1.3)

Result are good, the girl is always in the same position and we are able to improve the background, colors, etc.

Improve or change the position manually

If you click on the second edit it will open the "SD-WEBUI-OPENPOSE-EDITOR"

I decided to :

1) move up the yellow bone because it was an error in the image

2) Move down a little bit the hand

3) Put back 1 leg to the floor

Click on "Send pose to ControlNet" and

and now the image is much better

If you set the seed of the image you like, than you improve your prompt with negative prompt , se the "High rest" to some size and increase the "Sampling steps"

You can get a nicer image.

(From here he can improve hands, face ,feets with Adetailer component)

Saving and loading Position

You can use the buttons "Download Json" to save the current position

and "Upload Json" to load another position.

I have download the image, move the first one to the left, and imported again the same pose.

I send the pose to controlNet

And now I have 2 girls jumpin in similar positions.

Change the position

To select the many dots click in a place and move the mouse , the dots inside the rectangle will be selected.

When you have selected the dots you can resize all together, you can also mirror horizontally or vertically

And now we have the 2 girls

If I want to deferenziate the prompt you can use the "Regional Prompter", please read my article:

https://civitai.com/articles/3959/stable-diffusion-testing-the-regional-prompter-plug-in-using-break

Prompt

1woman blonde wearing (red bikini:1.2) . Jumping in the (beach:1.2) (full body:1.3)

BREAK

brunette little 1girl with school blue dress

A that is ..

You can see that Openpose is a very powerful component.. you can take any picture that you want.. a picture of yourself and the system will convert to a sketetron that you can adapt and save.. to then restore .. you can also have a collection of your favorites poses.

I hope you enjoy my article.

13

Comments