Mary-Cassatt-Oil-Flux-LoKr-Messier-Phase1-EMA-SS1_5-Log-SNR-FFS

This is a LyCORIS adapter derived from black-forest-labs/FLUX.1-dev.

No validation prompt was used during training.

None

Validation settings

• CFG: 3.0
• CFG Rescale: 0.0
• Steps: 20
• Sampler: FlowMatchEulerDiscreteScheduler
• Seed: 42
• Resolution: 968x1280
• 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:

Prompt

unconditional (blank prompt)

Negative Prompt

blurry, cropped, ugly

Prompt

In the style of a c4ss4tt oil painting, A child wearing an elaborate blue silk dress with ruffles and white lace trim sits near a window, the fabric catching soft light.

Negative Prompt

blurry, cropped, ugly

Prompt

In the style of a c4ss4tt oil painting, A close portrait of a young child's face with rosy cheeks and delicate features, gentle lighting from a nearby window.

Negative Prompt

blurry, cropped, ugly

Prompt

In the style of a c4ss4tt oil painting, Strong window light falls across a child's face and shoulder, creating bold shadows on their blue dress.

Negative Prompt

blurry, cropped, ugly

Prompt

In the style of a c4ss4tt oil painting, A child in a blue hat stands by a window.

Negative Prompt

blurry, cropped, ugly

Prompt

In the style of a c4ss4tt oil painting, A woman in soft colors holds her baby close.

Negative Prompt

blurry, cropped, ugly

Prompt

In the style of a c4ss4tt oil painting, A woman in a detailed white lace dress reads while seated by a window with gauzy curtains, various textures visible in the furnishings.

Negative Prompt

blurry, cropped, ugly

Prompt

In the style of a c4ss4tt oil painting, A mother in a textured knit sweater checks her phone while her baby sleeps against her shoulder.

Negative Prompt

blurry, cropped, ugly

Prompt

In the style of a c4ss4tt oil painting, A mother cat grooms her kitten by a sunlit window, their fur catching the light.

Negative Prompt

blurry, cropped, ugly

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

Training settings

• Training epochs: 1
• Training steps: 2900
• Learning rate: 0.0002
  • Learning rate schedule: polynomial
  • Warmup steps: 100
• Max grad norm: 0.1
• Effective batch size: 3
  • Micro-batch size: 3
  • Gradient accumulation steps: 1
  • Number of GPUs: 1
• Gradient checkpointing: True
• Prediction type: flow-matching (extra parameters=['shift=1.0', 'flux_guidance_mode=constant', 'flux_guidance_value=1.0', 'flow_matching_loss=compatible'])
• Optimizer: adamw_bf16
• Trainable parameter precision: Pure BF16
• Caption dropout probability: 10.0%

LyCORIS Config:

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

Datasets

cassatt-512

• Repeats: 22
• Total number of images: 50
• Total number of aspect buckets: 7
• Resolution: 0.262144 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

cassatt-768

• Repeats: 22
• Total number of images: 50
• Total number of aspect buckets: 8
• Resolution: 0.589824 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

cassatt-1024

• Repeats: 10
• Total number of images: 50
• Total number of aspect buckets: 11
• Resolution: 1.048576 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

cassatt-1536

• Repeats: 10
• Total number of images: 49
• Total number of aspect buckets: 13
• Resolution: 2.359296 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

cassatt-crops-512

• Repeats: 11
• Total number of images: 25
• Total number of aspect buckets: 6
• Resolution: 0.262144 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

cassatt-crops-768

• Repeats: 11
• Total number of images: 25
• Total number of aspect buckets: 11
• Resolution: 0.589824 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

cassatt-crops-1024

• Repeats: 5
• Total number of images: 25
• Total number of aspect buckets: 17
• Resolution: 1.048576 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

cassatt-crops-1536

• Repeats: 5
• Total number of images: 24
• Total number of aspect buckets: 17
• Resolution: 2.359296 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None
• 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 = 'black-forest-labs/FLUX.1-dev'
adapter_repo_id = 'davidrd123/Mary-Cassatt-Oil-Flux-LoKr-Messier-Phase1-EMA-SS1_5-Log-SNR-FFS'
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 = "An astronaut is riding a horse through the jungles of Thailand."


## 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,
    num_inference_steps=20,
    generator=torch.Generator(device='cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu').manual_seed(42),
    width=968,
    height=1280,
    guidance_scale=3.0,
).images[0]
image.save("output.png", format="PNG")

Exponential Moving Average (EMA)

SimpleTuner generates a safetensors variant of the EMA weights and a pt file.

The safetensors file is intended to be used for inference, and the pt file is for continuing finetuning.

The EMA model may provide a more well-rounded result, but typically will feel undertrained compared to the full model as it is a running decayed average of the model weights.