BitDiffusionV0.1

<Gallery /> ## BitDiffusionV0.1

This is the initial version of the image model trained on the Bittensor network within subnet 17. It's not expected for this model to perform as well as MidJourney V6 at the moment. However, it does generate better images than base SDXL model.

Trained on the dataset of Subnet 19 Vision.

Subnet 17 Checkpoint

Model ID : gtsru/sn17-dek-012

Revision : 5852d39e8413a377a3477b8278ade9af311f83a4

UID : 42

Perplexity : 1.1325

Settings for BitDiffusionV0.1

Use these settings for the best results with BitDiffusionV0.1:

CFG Scale: Use a CFG scale of 8

Steps: 40 to 60 steps

Sampler: DPM++ 2M SDE

Scheduler: Karras

Resolution: 1024x1024

For best results, set a negative_prompt

Use it with 🧨 diffusers

import torch
from diffusers import (
    StableDiffusionXLPipeline, 
    KDPM2AncestralDiscreteScheduler,
    AutoencoderKL
)

# Load VAE component
vae = AutoencoderKL.from_pretrained(
    "madebyollin/sdxl-vae-fp16-fix", 
    torch_dtype=torch.float16
)

# Configure the pipeline
pipe = StableDiffusionXLPipeline.from_pretrained(
    "PlixAI/BitDiffusionV0.1", 
    vae=vae,
    torch_dtype=torch.float16
)
pipe.scheduler = KDPM2AncestralDiscreteScheduler.from_config(pipe.scheduler.config)
pipe.to('cuda')

# Define prompts and generate image
prompt = "black fluffy gorgeous dangerous cat animal creature, large orange eyes, big fluffy ears, piercing gaze, full moon, dark ambiance, best quality, extremely detailed"
negative_prompt = "nsfw, bad quality, bad anatomy, worst quality, low quality, low resolutions, extra fingers, blur, blurry, ugly, wrongs proportions, watermark, image artifacts, lowres, ugly, jpeg artifacts, deformed, noisy image"

image = pipe(
    prompt, 
    negative_prompt=negative_prompt, 
    width=1024,
    height=1024,
    guidance_scale=7.5,
    num_inference_steps=50
).images[0]

Training Subnet : https://github.com/PlixML/pixel

Data Subnet : https://github.com/namoray/vision