Flux-Marsden-Hartley-LoKr-SimpleTuner-03

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

The main validation prompt used during training was:

hamster in the style of MRSDN

Validation settings

CFG: 3.0
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
unconditional (blank prompt)

Negative Prompt
blurry, cropped, ugly

Prompt
A rocky Maine coastline with bold, geometric shapes representing cliffs and waves. Strong colors and simplified forms dominate the composition, in the style of MRSDN

Negative Prompt
blurry, cropped, ugly

Prompt
An abstract composition inspired by Berlin's urban life. Fragmented shapes, numbers, and symbols arranged in a Cubist-influenced style, in the style of MRSDN

Negative Prompt
blurry, cropped, ugly

Prompt
A still life of flowers in a vase, rendered with thick brushstrokes and vibrant, non-naturalistic colors. Simplified forms show Cubist influence, in the style of MRSDN

Negative Prompt
blurry, cropped, ugly

Prompt
A stark New Mexico landscape with stylized mountains and desert flora. Bold outlines and earthy colors capture the essence of the Southwest, in the style of MRSDN

Negative Prompt
blurry, cropped, ugly

Prompt
A portrait of a WWI German soldier, composed of geometric shapes and military symbols. Strong, emotive use of color and form, in the style of MRSDN

Negative Prompt
blurry, cropped, ugly

Prompt
Mount Katahdin in Maine, depicted with sharp angles and bold colors. The landscape is reduced to its essential forms, emphasizing its rugged nature, in the style of MRSDN

Negative Prompt
blurry, cropped, ugly

Prompt
Modern New York skyscrapers rendered in Hartley's style. Geometric shapes and bold colors create a dynamic urban composition, in the style of MRSDN

Negative Prompt
blurry, cropped, ugly

Prompt
An orbiting space station viewed through a Modernist lens. Fragmented forms and symbolic elements represent the futuristic structure, in the style of MRSDN

Negative Prompt
blurry, cropped, ugly

Prompt
An electric car charging station, depicted with Cubist-inspired fragmentation. Bold colors and geometric shapes represent energy and technology, in the style of MRSDN

Negative Prompt
blurry, cropped, ugly

Prompt
A composition of social media icons and symbols, arranged in a Modernist style reminiscent of Hartley's German officer paintings, in the style of MRSDN

Negative Prompt
blurry, cropped, ugly

Prompt
An abstract representation of climate change, using Hartley's bold style to depict melting ice caps, rising seas, and changing weather patterns, in the style of MRSDN

Negative Prompt
blurry, cropped, ugly

Prompt
A person wearing a VR headset, surrounded by fragmented, Cubist-inspired virtual elements. Bold colors and geometric forms dominate the composition, in the style of MRSDN

Negative Prompt
blurry, cropped, ugly

Prompt
hamster, in the style of MRSDN

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: 12
Training steps: 7400
Learning rate: 0.0004
Effective batch size: 2
- Micro-batch size: 2
- 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

marsden-hartley-Flux-CC-512

Repeats: 10
Total number of images: 25
Total number of aspect buckets: 4
Resolution: 0.262144 megapixels
Cropped: False
Crop style: None
Crop aspect: None

marsden-hartley-Flux-CC-1024

Repeats: 10
Total number of images: 25
Total number of aspect buckets: 8
Resolution: 1.048576 megapixels
Cropped: False
Crop style: None
Crop aspect: None

marsden-hartley-Flux-CC-512-crop

Repeats: 10
Total number of images: 25
Total number of aspect buckets: 1
Resolution: 0.262144 megapixels
Cropped: True
Crop style: random
Crop aspect: square

marsden-hartley-Flux-CC-1024-crop

Repeats: 10
Total number of images: 25
Total number of aspect buckets: 1
Resolution: 1.048576 megapixels
Cropped: True
Crop style: random
Crop aspect: square

Inference

import torch
from diffusers import DiffusionPipeline
from lycoris import create_lycoris_from_weights

model_id = 'black-forest-labs/FLUX.1-dev'
adapter_id = 'pytorch_lora_weights.safetensors' # you will have to download this manually
lora_scale = 1.0
wrapper, _ = create_lycoris_from_weights(lora_scale, adapter_id, pipeline.transformer)
wrapper.merge_to()

prompt = "hamster in the style of MRSDN"

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=1024,
    height=1024,
    guidance_scale=3.0,
).images[0]
image.save("output.png", format="PNG")

davidrd123
/

Flux-Marsden-Hartley-LoKr-SimpleTuner-03