Meet jp-Amanda-400: My Latest Attempt in Textural Inversion Trainings

I'm thrilled to present jp-Amanda-400, another milestone in my exploration of textural inversion. Build on the Stable Diffusion 1.5 model.

Training Data:

• Trained from 22 images cropped and hand captioned cooked for 400 steps to ensure the right aesthetic but not to reproduce an actual person.

• Learning rate of 0.005

• Enhanced textural depth with 12 vectors per token.

• Batch size of 1 and 1 gradient accumulation step.

• Images used in creation of embedding skew towards NSFW so prompt accordingly.

I'm interested to see the diverse and imaginative uses of jp-Amanda-400. Your creative interpretations and shared artworks help keep me motivated. I invite you to experiment, share your results, and provide your valuable feedback.