sd35-training

sd35-training

This is a LyCORIS adapter derived from stabilityai/stable-diffusion-3.5-large.

The main validation prompt used during training was:

emaSde3Ver1, a high-resolution photograph featuring a young caucasian woman with long, wavy, platinum blonde hair cascading over her shoulders, she has a slender yet curvaceous physique with prominent breasts and a small waist, her skin is fair and smooth, with a slight blush on her cheeks, giving her a sultry expression, she is wearing a sheer, black fishnet bodysuit that accentuates her curves, with her back to the viewer, revealing her lower back and buttocks, the bodice is made of a soft, textured fabric that clings to her body, emphasizing her curves and the texture of the fishnet fabric, she also wears a black choker around her neck, adding a touch of sensuality to her attire, the background features a blurred, out-of-focus view of a cityscape with distant mountains and a clear blue sky, suggesting an outdoor setting, the balcony she is standing on has wooden railings and a wooden railing, adding to the sense of a balcony or terrace, the overall mood of the photograph is sensual and intimate, emphasizing the subject's allure and beauty

Validation settings

• CFG: 5.0
• CFG Rescale: 0.0
• Steps: 30
• Sampler: None
• Seed: 42
• Resolution: 1024

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: 139
• Training steps: 3500
• Learning rate: 0.0001
• Max grad norm: 0.01
• Effective batch size: 2
  • Micro-batch size: 2
  • Gradient accumulation steps: 1
  • Number of GPUs: 1
• Prediction type: flow-matching (extra parameters=['shift=3'])
• Rescaled betas zero SNR: False
• Optimizer: adamw_bf16
• Precision: Pure BF16
• Quantised: Yes: int8-quanto
• Xformers: Not used
• LyCORIS Config:

{
    "bypass_mode": true,
    "algo": "lokr",
    "multiplier": 1.0,
    "full_matrix": true,
    "linear_dim": 10000,
    "linear_alpha": 1,
    "factor": 12,
    "apply_preset": {
        "target_module": [
            "Attention"
        ],
        "module_algo_map": {
            "Attention": {
                "factor": 6
            }
        }
    }
}

Datasets

emaSde3Ver1

• Repeats: 0
• Total number of images: 50
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: true
• Crop style: center
• Crop aspect: square
• Used for regularisation data: No

Inference

import torch
from diffusers import DiffusionPipeline
from lycoris import create_lycoris_from_weights


def download_adapter(repo_id: str):
    import os
    from huggingface_hub import hf_hub_download
    adapter_filename = "pytorch_lora_weights.safetensors"
    cache_dir = os.environ.get('HF_PATH', os.path.expanduser('~/.cache/huggingface/hub/models'))
    cleaned_adapter_path = repo_id.replace("/", "_").replace("\\", "_").replace(":", "_")
    path_to_adapter = os.path.join(cache_dir, cleaned_adapter_path)
    path_to_adapter_file = os.path.join(path_to_adapter, adapter_filename)
    os.makedirs(path_to_adapter, exist_ok=True)
    hf_hub_download(
        repo_id=repo_id, filename=adapter_filename, local_dir=path_to_adapter
    )

    return path_to_adapter_file
    
model_id = 'stabilityai/stable-diffusion-3.5-large'
adapter_repo_id = 'alexnvo/sd35-training'
adapter_filename = 'pytorch_lora_weights.safetensors'
adapter_file_path = download_adapter(repo_id=adapter_repo_id)
pipeline = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.bfloat16) # loading directly in bf16
lora_scale = 1.0
wrapper, _ = create_lycoris_from_weights(lora_scale, adapter_file_path, pipeline.transformer)
wrapper.merge_to()

prompt = "emaSde3Ver1, a high-resolution photograph featuring a young caucasian woman with long, wavy, platinum blonde hair cascading over her shoulders, she has a slender yet curvaceous physique with prominent breasts and a small waist, her skin is fair and smooth, with a slight blush on her cheeks, giving her a sultry expression, she is wearing a sheer, black fishnet bodysuit that accentuates her curves, with her back to the viewer, revealing her lower back and buttocks, the bodice is made of a soft, textured fabric that clings to her body, emphasizing her curves and the texture of the fishnet fabric, she also wears a black choker around her neck, adding a touch of sensuality to her attire, the background features a blurred, out-of-focus view of a cityscape with distant mountains and a clear blue sky, suggesting an outdoor setting, the balcony she is standing on has wooden railings and a wooden railing, adding to the sense of a balcony or terrace, the overall mood of the photograph is sensual and intimate, emphasizing the subject's allure and beauty"
negative_prompt = 'blurry, cropped, ugly'

## Optional: quantise the model to save on vram.
## Note: The model was quantised during training, and so it is recommended to do the same 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(1641421826),
    width=1024,
    height=1024,
    guidance_scale=5.0,
).images[0]
image.save("output.png", format="PNG")