flux-lora-training

flux-lora-training

This is a standard PEFT LoRA derived from black-forest-labs/FLUX.1-dev.

The main validation prompt used during training was:

(ultra detailed,photo,photograph,best quality,high resolution,4k,8k,photorealistic,Japanese early twenties,(slim) and curvy body,waist,detailed beautiful eyes,super detailed eyes and skins,very beautiful woman:3.0), (Wearing tight short sleeves white (one piece) sailor uniform, blue collar, red neckerchief, dark blue pleated dress, standing in empty classroom,sweet smiling:2.0), long straight hair, standing in empty classroom,

Validation settings

• CFG: 3.5
• CFG Rescale: 0.0
• Steps: 20
• Sampler: FlowMatchEulerDiscreteScheduler
• Seed: 42
• Resolution: 1024x1024
• 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:

<Gallery />

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

Training settings

• Training epochs: 92

• Training steps: 2400

• Learning rate: 0.0001

  • Learning rate schedule: constant_with_warmup
  • Warmup steps: 100

• Max grad value: 1.0

• 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=['flow_schedule_auto_shift', 'shift=0.0', 'flux_guidance_mode=constant', 'flux_guidance_value=1.0', 'flow_matching_loss=compatible', 'flux_lora_target=all+ffs'])

• Optimizer: adamw_bf16

• Trainable parameter precision: Pure BF16

• Base model precision: no_change

• Caption dropout probability: 0.0%

• LoRA Rank: 16

• LoRA Alpha: None

• LoRA Dropout: 0.1

• LoRA initialisation style: default

Datasets

cissy

• Repeats: 0
• Total number of images: 26
• 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

model_id = 'black-forest-labs/FLUX.1-dev'
adapter_id = 'hok00i3/flux-lora-training'
pipeline = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.bfloat16) # loading directly in bf16
pipeline.load_lora_weights(adapter_id)

prompt = "(ultra detailed,photo,photograph,best quality,high resolution,4k,8k,photorealistic,Japanese early twenties,(slim) and curvy body,waist,detailed beautiful eyes,super detailed eyes and skins,very beautiful woman:3.0), (Wearing tight short sleeves white (one piece) sailor uniform, blue collar, red neckerchief, dark blue pleated dress, standing in empty classroom,sweet smiling:2.0), long straight hair, standing in empty classroom,"


## Optional: quantise the model to save on vram.
## Note: The model was not quantised during training, so it is not necessary to quantise it 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
model_output = 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=1024,
    height=1024,
    guidance_scale=3.5,
).images[0]

model_output.save("output.png", format="PNG")