flux2k9b-simpletuner-lora-loona

This is a PEFT LoRA derived from black-forest-labs/FLUX.2-klein-base-9B.

The main validation prompt used during training was:

Loona with her pale grey furand red eyes sits at a desk typing on a laptop with a focused exprssion, wearing her black studded outfit andchoker. The office has pink-tinted walls with a potted plant visible on the desk.

Validation settings

• CFG: 4.0
• CFG Rescale: 0.0
• Steps: 20
• Sampler: FlowMatchEulerDiscreteScheduler
• Seed: None
• Resolution: 1024x1024

Note: The validation settings are not necessarily the same as the training settings.

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

Training settings

• Training epochs: 24

• Training steps: 2500

• Learning rate: 1e-05

  • Learning rate schedule: cosine
  • Warmup steps: 100

• Max grad value: 2.0

• Effective batch size: 2

  • Micro-batch size: 1
  • Gradient accumulation steps: 2
  • Number of GPUs: 1

• Gradient checkpointing: True

• Prediction type: flow_matching[]

• Optimizer: optimi-lion

• Trainable parameter precision: Pure BF16

• Base model precision: int8-torchao

• Caption dropout probability: 0.1%

• LoRA Rank: 16

• LoRA Alpha: None

• LoRA Dropout: 0.1

• LoRA initialisation style: default

• LoRA mode: Standard

Datasets

loona-flux2

• Repeats: 3
• Total number of images: 50
• Total number of aspect buckets: 1
• Resolution: 0.262144 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

Inference

import torch
from diffusers import DiffusionPipeline

model_id = 'black-forest-labs/FLUX.2-klein-base-9B'
adapter_id = 'markury/flux2k9b-simpletuner-lora-loona'
pipeline = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.bfloat16) # loading directly in bf16
pipeline.load_lora_weights(adapter_id)

prompt = "Loona with her pale grey furand red eyes sits at a desk typing on a laptop with a focused exprssion, wearing her black studded outfit andchoker. The office has pink-tinted walls with a potted plant visible on the desk."
negative_prompt = 'blurry, cropped, ugly'

## 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
model_output = pipeline(
    prompt=prompt,
    negative_prompt=negative_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=4.0,
).images[0]

model_output.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.