killua7362 / animefusion Goto Github PK

AnimeFusion

Stable diffusion implementation with transformers!
Report Bug · Request Feature

Stable diffusion is a vital concept in the realm of generative models, serving as a technique to enhance the quality and diversity of generated data. It addresses key challenges associated with traditional generative models, such as mode collapse and limited variability in generated samples. At its core, stable diffusion involves a controlled process of transitioning data from an initial state to a final state while incrementally introducing noise at each step. This controlled noise injection allows generative models to explore a broader spectrum of data variations, resulting in more realistic and diverse output.

• Python
• Pytorch
• transformers
• diffusers
• numpy

Install all the libraries

pip install pytorch torchvision transformers diffusers numpy albumentations

Change the arguments value like dataset,batch size,etc in trainer.py file and call any function train or test at the end of the file and run trainer.py

• Implementing stable diffusion with Unet,VAE and clip model
• Writing both test and train function
• Training on Anime dataset
• Trying different hyperparameters

See the open issues for a full list of proposed features (and known issues).

• https://www.kaggle.com/datasets/shanmukh05/anime-names-and-image-generation
• https://arxiv.org/abs/2302.05543