The Ultimate Tutorial
First, install Stable Diffusion
Now we need a model to work with
https://civitai.com/api/download/models/29461?type=Model&format=PickleTensor&size=full&fp=fp16
Download the model from the link and paste it in
Stable diffusion\stable-diffusion-webui\models\Stable-diffusion
P.S. I’m using the 2.0 model but you can also try the 3.0 version.
Realistic Vision V3.0 - v3.0-inpainting (VAE) | Stable Diffusion Checkpoint | Civitai
Now before starting Stable diffusion, we’ll edit the webui-user.bat file
Right-click on it and edit
then add -xformers --autolaunch to COMMANDLINE_ARGS
It’ll look like
set COMMANDLINE_ARGS= --xformers --autolaunch
xformers will increase your image generation speed by around 40% and autolaunch is self-explanatory
and you can also add
git pull in a new line to get automatic updates from the git repo of Stable Diffusion
Open the webui-user.bat and it’ll install the necessary updates. It may take some time for the first time. So you need to be patient and let it run.
It’ll autolaunch the web interface in your default browser. If it doesn’t kindly copy the local server link from the webui command prompt and paste it into your browser
Now, we begin the editing
The interface will look like this.
- Here select the model in Stable Diffusion Checkpoint
- Then click on img2img (image to image) then click on inpaint
- Now fill up your prompts
- Then upload a pic and click on brush to paint the area which you wanna nudify
- Let the mask blur at around 4. It avoids shark edges in the inpainting area and also considers the area around the inpaint.
- Now scroll down and select inpaint area to only masked
- Now in sampling models, I prefer DPM++ SDE Karras, LMS Karras, and DPM++ 2s a. You can play around and can find what works for you best. You need to switch the methods if you’re not getting desired results
- Now CGF scale, so basically it give artistic freedom to the AI. So if you increase it the AI will have more of a say in how it wants the pic to be. You can play around with it and my desired range is between 7.5 and 8.5 and sometimes I tune it to 9-10 when I’m not getting my desired results
Now the seed,
Basically, it represents the parameters that we used to create the image. So in simple terms, if you got a banger image but need to make some little changes but want the pic to be the same then click on the green recycle button. It’ll save all your parameters and next time when you generate you’ll get the same pic with your additional changes.
Click on the dice to let it run randomly until you find your desired image or you can put -1 value in the box.
These are base prompts. What I do is I change around the prompts in every pic to get better results.
+ve Prompts (What you want in the pic)
RAW photograph, shot with Canon6D Mark 2, photorealistic, photograph of ((stunningly attractive)) (topless:1.6) kind eyes, ((delicate and small breasts)), ((naked breasts and navel and pussy)), legs spread apart , erect nipples, 20_degree_nipple_inclination, 40_40_breast_ratio, 23 yo, Indian actress, standing topless, busty petite, high definition breasts, highly detailed breasts, high definition nipples, highly detailed nipples, erect nipples, colour realistic photo, posing naked, embarrassed naked female short curly hair, depth of field, covered in piss, full body seen, Head to toe, fit nude woman, smooth pussy, stunningly beautiful, perfect feminine face, dat gap between the thighs, space above head, beautiful abs, full body, bright skin, (beautiful face), beautiful eyes, flirting into camera, full body, naked, phenomenal natural breasts, hyper realistic
-ve Prompts ( What you don’t want in the pic)
3d, cartoon, funny, unrealistic, symmetrical, deformed face, bad anatomy, extra legs, extra arms, poorly drawn face, poorly drawn hands, poorly drawn feet, disfigured, out of frame, long neck, bad hands, bad art, deformed, cross eyed, cgi, partial body, deformed, cross-eye, ((blurry)), ((bad anatomy)), ((ugly)), ((disfigured)), (poorly drawn face), mutation, mutated, extra limbs, extra nipples, (long neck), (ugly face), cloned face, unsymmetrical eyes, bad art, bad eyes, poorly drawn hands, poorly drawn fingers, red dress, clothes, deformation, Piercings
You can also supplement your edits with
LoRA models...
ControlNet is an additional advanced extension and we can talk about it another time...
You can find your outputs in Stable diffusion\stable-diffusion-webui\outputs\img2img-images
Here you go, guys.
Let’s see who can surpass their sensei…
Thank You
AamRush