pixart-900m-1024-ft-v0.7-stage2

pixart-900m-1024-ft-v0.7-stage2

This is a full rank finetune derived from terminusresearch/pixart-900m-1024-ft-v0.6.

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:

<Gallery />

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

Training settings

• Training epochs: 9
• Training steps: 29500
• Learning rate: 1e-06
• Effective batch size: 16
  • Micro-batch size: 16
  • Gradient accumulation steps: 1
  • Number of GPUs: 1
• Prediction type: epsilon
• Rescaled betas zero SNR: False
• Optimizer: AdamW, stochastic bf16
• Precision: Pure BF16
• Xformers: Enabled

Datasets

shutterstock

• Repeats: 0
• Total number of images: 21040
• Total number of aspect buckets: 3
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: random

nijijourney

• Repeats: 0
• Total number of images: 21488
• 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: 89296
• 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: 33120
• Total number of aspect buckets: 3
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: random

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

cinemamix-1mp

• Repeats: 0
• Total number of images: 7376
• Total number of aspect buckets: 5
• Resolution: 1.0 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None

Inference

import torch
from diffusers import DiffusionPipeline

model_id = 'pixart-900m-1024-ft-v0.7-stage2'
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")