SD3.5-LoRA-Korean-Basic-Design

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

The main validation prompt used during training was:

Using a strawberry and a stainless steel whisk, create a composition that expresses a sense of rhythm and vibrancy, basic design.

Validation settings

CFG: 5.0
CFG Rescale: 0.0
Steps: 20
Sampler: None
Seed: 42
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:

Prompt
Using a strawberry and a stainless steel whisk, create a composition that expresses a sense of rhythm and vibrancy, basic design.

Prompt
A dynamic interplay of tumbling dice and flowing red ribbons wrapping around metallic pipes, basic design style.

Prompt
A dynamic composition of transparent light bulbs intertwined with flowing golden and blue ribbons, basic design style.

Prompt
Using a wine glass and wooden tongs, express a moment of tension and destruction, basic design.

Prompt
A dynamic explosion of metallic whistles and shattering biscuit sticks, all entangled with vibrant, colorful ribbons, basic design style.

Prompt
A glass chess piece being tightly wrapped and constricted by a heavy, metallic chain, creating a sense of tension and imminent fracture, basic design.

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

Training settings

Training epochs: 2
Training steps: 2000
Learning rate: 0.0001
Max grad norm: 0.01
Effective batch size: 1
- Micro-batch size: 1
- Gradient accumulation steps: 1
- Number of GPUs: 1
Prediction type: flow-matching
Rescaled betas zero SNR: False
Optimizer: adamw_bf16
Precision: Pure BF16
Quantised: Yes: int8-quanto
Xformers: Not used
LyCORIS Config:

{
    "algo": "lokr",
    "multiplier": 1.0,
    "linear_dim": 10000,
    "linear_alpha": 1,
    "factor": 16,
    "apply_preset": {
        "target_module": [
            "Attention",
            "FeedForward"
        ],
        "module_algo_map": {
            "Attention": {
                "factor": 16
            },
            "FeedForward": {
                "factor": 8
            }
        }
    }
}

Datasets

my-dataset-512

Repeats: 10
Total number of images: 17
Total number of aspect buckets: 1
Resolution: 0.262144 megapixels
Cropped: False
Crop style: None
Crop aspect: None
Used for regularisation data: No

my-dataset-1024

Repeats: 10
Total number of images: 17
Total number of aspect buckets: 4
Resolution: 1.048576 megapixels
Cropped: False
Crop style: None
Crop aspect: None
Used for regularisation data: No

my-dataset-512-crop

Repeats: 10
Total number of images: 17
Total number of aspect buckets: 1
Resolution: 0.262144 megapixels
Cropped: True
Crop style: random
Crop aspect: square
Used for regularisation data: No

my-dataset-1024-crop

Repeats: 10
Total number of images: 17
Total number of aspect buckets: 1
Resolution: 1.048576 megapixels
Cropped: True
Crop style: random
Crop aspect: square
Used for regularisation data: No

Inference

import torch
from diffusers import StableDiffusion3Pipeline
from lycoris import create_lycoris_from_weights

adapter_id = 'taewan2002/SD3.5-LoRA-Korean-Basic-Design'
pipeline = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3.5-large", torch_dtype=torch.bfloat16)
lora_scale = 1.0
wrapper, _ = create_lycoris_from_weights(lora_scale, adapter_id, pipeline.transformer)
wrapper.merge_to()

prompt = "Using a wine glass and wooden tongs, express a moment of tension and destruction, basic design."

pipeline.to('cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu')
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(1641421826),
    width=1072,
    height=720,
    guidance_scale=5.0,
).images[0]
image.save("output.png", format="PNG")

taewan2002
/

SD3.5-LoRA-Korean-Basic-Design