Extract Clothes LoRA

Inspiration

While doing virtual try-on a clothes that is worn by a person, I was unable to put that clothes directly to my model photo. The try on images are not looking perfect in that case. So, I feel a need to extract the clothes in plain background first. There was no AI tool I found that was able to do my this specific task perfectly. So, this inspires me to train this lora.

What it does

This lora extract clothes from a person photo. It extract clothes, watch, cap and put it in a plain background and product photography style.

How we built it

I used FLUX.1 Kontext [dev] as base model. We trained a specialized LoRA. For lora training, we made a special dataset for this. The training dataset pairs two things. Input of pair is, person photo worn a clothing where output of pair is only the clothing in plain background.

Challenges we ran into

The main challenges I face is dataset creation for this needs. It was hard to make the dataset for my specific needs.

Accomplishments that we're proud of

I am proud that I trains a lora, that is useful and helpful for various tasks like product photography, virtual try on etc. This lora will help in many workflows.

What we learned

This project was a deep dive into the creative potential of the FLUX.1 Kontext [dev] model. I learned that fine-tuning a custom LoRA can unlock entirely new abilities in a model.

What's next for Extract Clothes LoRA

I am going to make a AI image generator and photo editor SaaS with numerous options. So, I can add very good features on my SaaS and explore more possibility.

Built With

• fal
• huggingface
• lora-training
• python