hwasan-yc-tag

hwasan-yc-tag

This is a standard PEFT LoRA derived from stabilityai/stable-diffusion-3.5-large.

The main validation prompt used during training was:

k4s4, [speech-bubble-2] [people-2] [panel-1] [background-undefined] [camera-medium-shot] The scene depicts two characters in a heated exchange, with one character appearing visibly distressed or angry. They are engaged in a conversation, as indicated by the speech bubbles. The background is not clearly defined, suggesting an interior space, possibly a room with limited visibility of details. The shot captures both characters from a medium distance, emphasizing their expressions and the intensity of the moment.

Validation settings

• CFG: 7.5
• CFG Rescale: 0.0
• Steps: 30
• Sampler: FlowMatchEulerDiscreteScheduler
• Seed: 42
• Resolution: 512x512
• Skip-layer guidance:

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: 6

• Training steps: 3400

• Learning rate: 1e-05

  • Learning rate schedule: cosine
  • Warmup steps: 2500

• Max grad norm: 2.0

• Effective batch size: 8

  • Micro-batch size: 8
  • Gradient accumulation steps: 1
  • Number of GPUs: 1

• Gradient checkpointing: True

• Prediction type: flow-matching (extra parameters=['shift=3'])

• Optimizer: adamw_bf16

• Trainable parameter precision: Pure BF16

• Caption dropout probability: 20.0%

• LoRA Rank: 16

• LoRA Alpha: None

• LoRA Dropout: 0.1

• LoRA initialisation style: default

Datasets

webtoon-storyboard

• Repeats: 2
• Total number of images: 1383
• Total number of aspect buckets: 1
• Resolution: 0.262144 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

Inference

import torch
from diffusers import DiffusionPipeline

model_id = 'stabilityai/stable-diffusion-3.5-large'
adapter_id = 'gunchoi/hwasan-yc-tag'
pipeline = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.bfloat16) # loading directly in bf16
pipeline.load_lora_weights(adapter_id)

prompt = "k4s4, [speech-bubble-2] [people-2] [panel-1] [background-undefined] [camera-medium-shot] The scene depicts two characters in a heated exchange, with one character appearing visibly distressed or angry. They are engaged in a conversation, as indicated by the speech bubbles. The background is not clearly defined, suggesting an interior space, possibly a room with limited visibility of details. The shot captures both characters from a medium distance, emphasizing their expressions and the intensity of the moment."
negative_prompt = 'blurry, cropped, ugly'

## Optional: quantise the model to save on vram.
## Note: The model was not quantised during training, so it is not necessary to quantise it during inference time.
#from optimum.quanto import quantize, freeze, qint8
#quantize(pipeline.transformer, weights=qint8)
#freeze(pipeline.transformer)
    
pipeline.to('cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu') # the pipeline is already in its target precision level
image = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    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(42),
    width=512,
    height=512,
    guidance_scale=7.5,
).images[0]
image.save("output.png", format="PNG")