|

How To Change Clothes In Stable Diffusion [Guide]

I’ve covered a ton of prompts, models, and tutorials on this blog but one of the biggest use cases of Stable Diffusion or any AI art generator tool is changing clothes. 

Whether you want to turn your boring photos into something cool or want a new professional profile picture for LinkedIn, you can do so easily in Stable Diffusion.

I’ve seen many ads on social media about such tools that let you change clothes and they all are paid which is what inspired me to write this tutorial. 

Today, I’ll show you how to change clothes in Stable Diffusion so that you don’t have to spend money on such apps and tools. 

Let’s get started. 

Pre-requisites

Before we begin, you need to know everything that’s required to change clothes in Stable Diffusion. 

The first thing you need is Automatic1111 installed on your device which is a GUI for running Stable Diffusion. You can also use ComfyUI but I’d be using Automatic1111 for this tutorial. 

Then you’ll need to install the ControlNet extension in Automatic1111 which will allow you to use ControlNet models. We’ll be using the OpenPose ControlNet model for changing clothes. 

Lastly, you’ll need an inpainting checkpoint model as we’ll be doing img2img inpainting and normal checkpoint models won’t work well with that. You can choose any of the models I’ve recommended above. 

Once you have all this, you can begin by changing clothes in Stable Diffusion. 

How To Change Clothes In Stable Diffusion

As I mentioned earlier, we’ll be using the Inpainting feature found in the img2img tab of Automatic1111. 

With this feature, you basically paint a mask over an area and use prompts to modify or change it. So, we’ll be masking over the clothes of our chosen image and then customize it with some prompts. 

Let’s see how it’s done. 

In Automatic1111, go to img2img and click on the Inpaint tab. 

Automatic1111 - Img2Img Inpaint

Here, upload the image on which you wish to change the clothes. I’ll be using this image for the tutorial. 

Change Clothes In Stable Diffusion - Inpaint Input

Then, paint over the entire clothes in the image and make sure you leave out nothing. 

Change Clothes In Stable Diffusion - Inpaint Mask

Now, it’s time to write some prompts for changing the clothes. I want to change the clothes to something sci-fi or cyberpunk. 

So, here are the positive and negative prompts I’ve used: 

Positive Prompt: 

wearing a professional suit with a tie, female, (1girl), high heels, black pants

Negative Prompt: 

(monochrome:1.3), (deformed, distorted, disfigured:1.3), (hair), jeans, tattoo, wet, water, clothing, shadow, 3d render, cartoon, ((blurry)), duplicate, ((duplicate body parts)), (disfigured), (poorly drawn), ((missing limbs)), logo, signature, text, words, low res, boring, artifacts, bad art, gross, ugly, poor quality, low quality, poorly drawn, bad anatomy, wrong anatomy​, nsfw, nude

Once you’ve added your prompts, scroll down to the configuration settings and set the following options: 

  • Sampler: Euler A 
  • Sampling Steps: 30-35 
  • CFG: 7
  • Denoising Strength: 0.5-0.7
  • Inpaint Area: Only Masked
  • Image Size: 512×768
  • Mask Blur: 4-8
  • Resize Mode: Crop & Resize

You don’t have to change the rest of the settings and leave them as is. 

Now, choose your checkpoint model. Make sure you’re using an inpainting model and not a regular checkpoint model. I’m using the Clarity Inpaint checkpoint for this tutorial.

Click on the Generate button and let’s change the clothes. 

Change Clothes In Stable Diffusion - Inpaint Result Without ControlNet

As you can see, we’ve successfully changed the clothes in our image. That’s how easy it is to change clothes in Stable Diffusion. 

But oftentimes, changing clothes using this method causes problems. You’ll encounter situations where the pose of the person is completely changed. 

For example, when generating the above example, I encountered a result where the pose was completely changed. 

This could result in unwanted results and you’ll end up generating the image multiple times. 

That’s where ControlNet comes to the rescue. With the OpenPose model of ControlNet, we can detect the pose of the human and use that specific pose in our output image.

This way, the pose isn’t changed and you get better outputs. Let’s generate the same image we generated but using ControlNet. 

Paint over the entire clothes in the Inpaint area and select the same generation settings as you did before. 

Scroll down and you’ll find a ControlNet toggle. Expand the toggle and enable ControlNet. 

Change Clothes In Stable Diffusion - ControlNet Settings

Select the OpenPose checkbox and choose the OpenPose model in the Model dropdown. 

Also, enable the Low VRAM and Pixel Perfect checkboxes. 

That’s all you have to do. We don’t need to change any other setting here. 

Click on the Generate button and this time, our image will be generated using ControlNet. 

Here’s the output image compared to the one generated with just inpainting: 

As you can see, by using ControlNet we were able to preserve the pose of the human. 

This is why using the ControlNet OpenPose model is extremely helpful whenever you want to change clothes in Stable Diffusion. 

You can also see the pose detected by the OpenPose model in the output:

Change Clothes In Stable Diffusion - ControlNet OpenPose

Sure you can do it without ControlNet but that’s more of a hit-and-trial method. With ControlNet, you can be sure that the human pose will not be changed. 

Prompt Ideas for Changing Clothes In Stable Diffusion 

Now that you’ve learned how to change clothes in Stable Diffusion, let’s take a look at some prompt ideas you can experiment with. 

The ability to change clothes in Stable Diffusion is only useful if you can use it in real-world applications or just to explore different art styles. 

So, here are some prompt ideas you can try out. 

Professional Headshots for LinkedIn 

(1guy), wearing a suit with a black blazer and blue tie, a professional headshot

Turn Yourself Into A Fashion Model

Victorian Style

(1girl), fashion model, Victorian style, wearing a white gown, elegant, beautiful 

Summer Style

(1guy), fashion model, summer style, Hawaiin shorts, colorful shirt, open buttons, wearing a necklace

Egyptian Style

sharp focus, realistic, 1girl, earrings, jewelry, lips, realistic, wearing Egyptian style dress, queen, luxurious 

Flower Style

(1girl), wearing a dress made of red flowers, intricate details, highly detailed, rose petals, beautiful, elegant 

Wedding Style

(1girl), wearing a white color wedding dress, bride, intricate details, beautiful, elegant 

You can also discover many clothes and dress models on Civitai to discover more ways you can change your clothes in Stable Diffusion. 

Conclusion

With Stable Diffusion, you can easily change your clothes and turn yourself into a rockstar, chef, fashion model, or whatever you want. 

I hope this tutorial helped you learn how to change clothes in Stable Diffusion. With this method, you can now ditch those tools and websites that cost you money and also could be a privacy nightmare. 

If you have any questions regarding this tutorial, let me know in the comments below.

Stable Diffusion Prompt Organizer

2 Comments

  1. Hi, congratulations for starting the stable diffusion journey.
    I need help or guide related to Stable Diffusion.
    How we change the clothes in stable diffusion automatically do inpainting process? Is there any other way.

    I know people says Segmentation anything or dino segmentation But This process is difficult for me to do this in stable diffusion.

    Can you guide me how to do this.
    Our task is only change the clothes of person.
    But in mobile application inpainting is not a good approach.

    ThankYou so much for your time.
    Regards
    M.Awais

    1. In mobile, you’d have to rely on other apps and websites that automatically do it for you. But if you are using a computer, then using inpainting in Stable Diffusion is the best approach for it.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.