maver1chh / cha2102_loha

cha2102_loha

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

The main validation prompt used during training was:

A peaceful Japanese-inspired scene unfolds, showcasing a cozy retreat nestled in the heart of nature. Towering mountains rise in the distance, framing a serene environment filled with vibrant plants and lush greenery. A calm pond reflects the bright sunlight, its surface adorned with delicate ripples and blooming lotus flowers\u2014where Frog basks on a lily pad, quietly observing the tranquil surroundings. Nearby, a rose garden adds a touch of romance, its soft petals contrasting beautifully with the earthy tones of the environment. Inside the rustic cottage, m0n1t0rs sitting in calm wearing headphones, adding a hint of nostalgic charm that complements the timeless beauty outside. This setting exudes tranquility, inviting you to pause, breathe, and connect with the harmony of nature\u2014a perfect haven where the natural splendor of Japan landscapes meets cozy serenity.

Validation settings

• CFG: 3.0
• CFG Rescale: 0.0
• Steps: 20
• Sampler: FlowMatchEulerDiscreteScheduler
• Seed: 42
• Resolution: 1344x768
• 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:

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

Training settings

• Training epochs: 20
• Training steps: 8000
• Learning rate: 0.0004
  • Learning rate schedule: polynomial
  • Warmup steps: 250
• Max grad norm: 2.0
• Effective batch size: 1
  • Micro-batch size: 1
  • Gradient accumulation steps: 1
  • Number of GPUs: 1
• Gradient checkpointing: True
• Prediction type: flow-matching (extra parameters=['shift=3', '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": "loha",
    "multiplier": 1.0,
    "linear_dim": 16,
    "linear_alpha": 16,
    "apply_preset": {
        "target_module": [
            "Attention",
            "FeedForward"
        ],
        "module_algo_map": {
            "Attention": {
                "factor": 16
            },
            "FeedForward": {
                "factor": 8
            }
        }
    }
}

Datasets

cha_2102_512

• Repeats: 5
• Total number of images: 32
• Total number of aspect buckets: 1
• Resolution: 0.262144 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

cha_2102_768

• Repeats: 5
• Total number of images: 32
• Total number of aspect buckets: 1
• Resolution: 0.589824 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 = 'maver1chh/cha2102_loha'
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 = "A peaceful Japanese-inspired scene unfolds, showcasing a cozy retreat nestled in the heart of nature. Towering mountains rise in the distance, framing a serene environment filled with vibrant plants and lush greenery. A calm pond reflects the bright sunlight, its surface adorned with delicate ripples and blooming lotus flowers\u2014where Frog basks on a lily pad, quietly observing the tranquil surroundings. Nearby, a rose garden adds a touch of romance, its soft petals contrasting beautifully with the earthy tones of the environment. Inside the rustic cottage, m0n1t0rs sitting in calm wearing headphones, adding a hint of nostalgic charm that complements the timeless beauty outside. This setting exudes tranquility, inviting you to pause, breathe, and connect with the harmony of nature\u2014a perfect haven where the natural splendor of Japan landscapes meets cozy serenity."


## 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=1344,
    height=768,
    guidance_scale=3.0,
).images[0]
image.save("output.png", format="PNG")