This tutorial is on the easy way to use X/Y plots to test things in Stable Diffusion. This is by far the best trick in the Stable Diffusion and must know.
With all the updates and 1000's of settings it is important to automate testing to find that sweet spot with the seed and CGF scale. This is how you do it. With X/Y plots.
X/Y plots can be found at the bottom of txt2img or img2img in a section called scripts.
What this tool does is it compares 2 values and puts one on an X (row) axis and the other value on a Y (column axis)
For example, you can choose Seed and CFG Scale to test which seeds work best with certain CFG's and test 100's of photo's while your AFK.
This is important to do every time Stable diffusion has an update or you downloaded a new ckpt file. You need to know what the sweet spot is. You can compare any 2 results.
You can also put nothing on the one of the plots and just test a single value over and over for example CFG 5 to CFG 12.
Let's test that first.
In the example bellow we are putting Seeds in the X value that range from 25 to 50 with 4 variations. Simply put, it will put a number that fits 4 times in the range of 25 to 50 starting with 25 and ending with 50.
The Y values with have the CFG scale from the range 6-10 (+0.5) this means that it will give me a variation every 0.5 increments from 5 until I reach 10, so 6.5, 7.0, 7.5 and so on...
Make Sure to leave a space between the range i.e. 25-50 and the variation or equation, i.e. (+0.5) or [4]
Here is a few of those examples:
prompt = j3nna0rt3ga showing peace sign at white sand beach in Hawaii north shore waves in background with coconut trees and tropical flowers, wearing beautiful flower dress, bokeh, professional color grading
Negative Prompt: ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, extra limbs, disfigured, deformed, body out of frame, blurry, bad anatomy, blurred, watermark, grainy, signature, cut off, draft, dim, mutated, badly drawn, dim, headless, no head, out of frame, malformed hands, missing limb
j3nna0rt3ga is a embedding file. To use these just go to this site or hugging face and search embeddings or textural inversion.
This site has a filter for them: Civitai | Stable Diffusion models, embeddings, hypernetworks and more
The are tiny files less than 1 MB normally and you can just drop them in your stable diffusion embeddings folder and you will have access to them.
My embeddings folder is C:\Users\PcUser\Documents\Stable Diffusion\stable-diffusion-webui\embeddings
Once you have the embedding in the folder just click on pink icon that looks like an old school iPod (young people might have to google that)
This button
Now it will show your embeddings and you can just click on to bring it in your prompt.
You can manually type them in but the names have to be exactly the same. It would help to change the file name of the embeddings in your stable diffusion folder.
Click refresh if you added the embeddings after you started stable diffusion.
I am also using a custom checkpoint file from the same site. Civitai | Stable Diffusion models, embeddings, hypernetworks and more
Here are some examples of the output.
The hands are messed up but fortunately for me that's what I want. My next tutorial will be how to fix messed up hands withing the control net depth mapping tool.
The cool thing is I could of tested 1,000 pictures comparing 2 or 3 settings in Stable Diffusion which makes a huge difference and find that perfect sweet spot for a new ckpt, embeddings or version of Stable Diffusion. This is a must know trick!!!!
So just set up the simple formula and then walk away for 10 min and come back to 100's of pictures.
By the way you can grab the seed from the file name
If your into Stable Diffusion, AI art or AI tools like Chat GPT and Voice Cloning I would recommend checking out my YouTube channel starting with this video below about Stable Diffusion Control net.
I will have a video for this x/y plot method in a bit.
Last tip!
Comments