Unconditioned stable diffusion finetuning - yurman/uncond-sd2-base-complex-4

This pipeline was finetuned from yurman/uncond_sd2-base for brain image generation. Below are some example images generated with the finetuned pipeline:

Pipeline usage

You can use the pipeline like so:

from diffusers import StableDiffusionUnconditionalPipeline
import torch

pipeline = StableDiffusionUnconditionalPipeline.from_pretrained("yurman/uncond-sd2-base-complex-4", torch_dtype=torch.float32)
image = pipeline(1).images[0]
image.save("brain_image.png")

Training info

These are the key hyperparameters used during training:

• Epochs: 400
• Max Train Steps: 100000
• Learning rate: 5e-05
• Batch size: 18
• VAE scaling: 0.12
• VAE type: MEDVAE
• Input perturbation: 0.0
• Noise offset: 0.0
• Gradient accumulation steps: 3
• Image resolution: 256
• Mixed-precision: no
• Max rotation degree: 10
• Prediction Type: v_prediction
• SNR Gamma: 5.0

More information on all the CLI arguments and the environment are available on your wandb run page.