This is a test model created to assess the Waifu Diffusion training code, and not intended to be a full-featured or official release.
This model has been trained from runwayml/stable-diffusion-v1-5 for approximately 1.6 epochs on 1.2m images total from various Instagram accounts (primarily Japanese). As the model is undertrained, its performance is marginal. Mixing the model is recommended for better performance.
Natural language descriptions (using BLIP), as well as booru tags have been used to assist in captioning. Any Instagram hashtags were also included in the caption data.
Note: Training was done using various aspect ratios, with a base resolution of 768x768, as well as the penultimate CLIP layer. Clip skip of 2 and a resolution of 768x768 or higher is recommended for generations.
