pixart-900m-1024-ft-v0.7-stage1

This is a full rank finetune derived from ptx0/pixart-900m-1024-ft-v0.7-stage1.

The main validation prompt used during training was:

a cute anime character named toast, holding a sign that reads SOON

Validation settings

• CFG: 4.0
• CFG Rescale: 0.7
• Steps: 30
• Sampler: None
• Seed: 420420420
• Resolution: 1024x1024

Note: The validation settings are not necessarily the same as the training settings.

You can find some example images in the following gallery:

The text encoder was not trained. You may reuse the base model text encoder for inference.

Training settings

• Training epochs: 6
• Training steps: 220000
• Learning rate: 1e-06
• Effective batch size: 64
  • Micro-batch size: 8
  • Gradient accumulation steps: 1
  • Number of GPUs: 8
• Prediction type: epsilon
• Rescaled betas zero SNR: False
• Optimizer: AdamW, stochastic bf16
• Precision: Pure BF16
• Xformers: Enabled

Datasets

sports

• Repeats: 0
• Total number of images: ~768
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

mj-60

• Repeats: 0
• Total number of images: ~179136
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

id-75k

• Repeats: 0
• Total number of images: ~36224
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

celebrities

• Repeats: 0
• Total number of images: ~1088
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

normalnudes

• Repeats: 0
• Total number of images: ~1024
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

guys

• Repeats: 0
• Total number of images: ~320
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

signs

• Repeats: 0
• Total number of images: ~320
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

dalle3

• Repeats: 0
• Total number of images: ~1112704
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

sfwbooru

• Repeats: 0
• Total number of images: ~395136
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

moviecollection

• Repeats: 0
• Total number of images: ~1792
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

bookcovers

• Repeats: 0
• Total number of images: ~704
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

nijijourney

• Repeats: 0
• Total number of images: ~512
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

experimental

• Repeats: 0
• Total number of images: ~3008
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

ethnic

• Repeats: 0
• Total number of images: ~3072
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

gay

• Repeats: 0
• Total number of images: ~1088
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

architecture

• Repeats: 0
• Total number of images: ~4352
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

shutterstock

• Repeats: 0
• Total number of images: ~21056
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

midjourney-v6-520k-raw

• Repeats: 0
• Total number of images: ~390976
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

nijijourney-v6-520k-raw

• Repeats: 0
• Total number of images: ~416064
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

cinemamix-1mp

• Repeats: 0
• Total number of images: ~7232
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

nsfw-1024

• Repeats: 0
• Total number of images: ~10752
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

anatomy

• Repeats: 5
• Total number of images: ~16384
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

bg20k-1024

• Repeats: 0
• Total number of images: ~89280
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

yoga

• Repeats: 0
• Total number of images: ~3584
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

photo-aesthetics

• Repeats: 0
• Total number of images: ~33088
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

text-1mp

• Repeats: 5
• Total number of images: ~13184
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

Inference

import torch
from diffusers import DiffusionPipeline

model_id = 'pixart-900m-1024-ft-v0.7-stage1'
pipeline = DiffusionPipeline.from_pretrained(model_id)

prompt = "a cute anime character named toast, holding a sign that reads SOON"
negative_prompt = "blurry, cropped, ugly"

pipeline.to('cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu')
image = pipeline(
    prompt=prompt,
    negative_prompt='blurry, cropped, ugly',
    num_inference_steps=30,
    generator=torch.Generator(device='cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu').manual_seed(1641421826),
    width=1152,
    height=768,
    guidance_scale=4.0,
    guidance_rescale=0.7,
).images[0]
image.save("output.png", format="PNG")