We found images of BMW M4 model online, and then cropped their dimensions to 512 x 512 px in bulk. We made sure the images we found have a clean background so that noises are minimized.

We used the LoRA (Low-Rank Adaption) method within DreamBooth to train Stable Diffusion as it allows us to just use a few images to generate new outputs. We decided to use LoRA because it allows faster training and image generation.

With the trained model from LoRA, we then cropped a new image and uploaded on a public link for us to extract it from the internet. Combining the new image and the text prompt, we were able to generate a newly upscaled images.

Built With

  • dreambooth
  • google-colab
  • huggingface
  • python
  • stable-diffusion
Share this project:

Updates