flux-sfwbooru-3.5M-lokr-attempt2

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

The main validation prompt used during training was:

A photo-realistic image of a cat

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 garfield waifu wearing an apron with a red sphere over her head that reads It is Time

Negative Prompt
blurry, cropped, ugly

Prompt
a void of fursuit furries hanging onto the edge of reality as they get sucked into a vortex

Negative Prompt
blurry, cropped, ugly

Prompt
furries congregate at walmart to teach about gelatin fountains to adult furries

Negative Prompt
blurry, cropped, ugly

Prompt
the furry church congregation looking up at a cinematic movie screen with text on it that reads MOOSE = PONY

Negative Prompt
blurry, cropped, ugly

Prompt
furry church congregation singing hymns while they look to a screen with lyrics on it that reads THE NEW FRONTIER OF PONY MODELS?

Negative Prompt
blurry, cropped, ugly

Prompt
a furry giving a TED talk with a screen in the background showing bullet points: - what furry means, and, - what furry does not mean

Negative Prompt
blurry, cropped, ugly

Prompt
a sugar bear named brownie plays basketball with lumps of poop

Negative Prompt
blurry, cropped, ugly

Prompt
A photo-realistic image of a cat

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: 0
Training steps: 3300
Learning rate: 0.0003
Effective batch size: 6
- Micro-batch size: 2
- Gradient accumulation steps: 1
- Number of GPUs: 3
Prediction type: flow-matching
Rescaled betas zero SNR: False
Optimizer: adamw_bf16weight_decay=1e-3
Precision: Pure BF16
Quantised: Yes: int8-quanto
Xformers: Not used
LyCORIS Config:

{
    "algo": "lokr",
    "multiplier": 1.0,
    "linear_dim": 10000,
    "linear_alpha": 1,
    "factor": 8,
    "apply_preset": {
        "target_module": [
            "Attention",
            "FeedForward"
        ],
        "module_algo_map": {
            "Attention": {
                "factor": 6
            },
            "FeedForward": {
                "factor": 4
            }
        }
    }
}

Datasets

sfwbooru

Repeats: 0
Total number of images: ~639264
Total number of aspect buckets: 80
Resolution: 1.048576 megapixels
Cropped: False
Crop style: None
Crop aspect: None

normalnudes

Repeats: 50
Total number of images: ~1233
Total number of aspect buckets: 30
Resolution: 1.048576 megapixels
Cropped: False
Crop style: None
Crop aspect: None

guys

Repeats: 0
Total number of images: ~390
Total number of aspect buckets: 16
Resolution: 1.048576 megapixels
Cropped: False
Crop style: None
Crop aspect: None

signs

Repeats: 0
Total number of images: ~435
Total number of aspect buckets: 19
Resolution: 1.048576 megapixels
Cropped: False
Crop style: None
Crop aspect: None

nsfw-1024

Repeats: 0
Total number of images: ~10830
Total number of aspect buckets: 14
Resolution: 1.048576 megapixels
Cropped: False
Crop style: None
Crop aspect: None

anatomy

Repeats: 2
Total number of images: ~16440
Total number of aspect buckets: 24
Resolution: 1.048576 megapixels
Cropped: False
Crop style: None
Crop aspect: None

yoga

Repeats: 0
Total number of images: ~3642
Total number of aspect buckets: 20
Resolution: 1.048576 megapixels
Cropped: False
Crop style: None
Crop aspect: None

photo-aesthetics

Repeats: 0
Total number of images: ~33171
Total number of aspect buckets: 30
Resolution: 1.048576 megapixels
Cropped: False
Crop style: None
Crop aspect: None

text-1mp

Repeats: 125
Total number of images: ~13221
Total number of aspect buckets: 28
Resolution: 1.048576 megapixels
Cropped: False
Crop style: None
Crop aspect: None

normalnudes-crop

Repeats: 50
Total number of images: ~1146
Total number of aspect buckets: 1
Resolution: 1.048576 megapixels
Cropped: True
Crop style: random
Crop aspect: square

guys-crop

Repeats: 0
Total number of images: ~381
Total number of aspect buckets: 1
Resolution: 1.048576 megapixels
Cropped: True
Crop style: random
Crop aspect: square

signs-crop

Repeats: 0
Total number of images: ~417
Total number of aspect buckets: 1
Resolution: 1.048576 megapixels
Cropped: True
Crop style: random
Crop aspect: square

nsfw-1024-crop

Repeats: 0
Total number of images: ~10818
Total number of aspect buckets: 1
Resolution: 1.048576 megapixels
Cropped: True
Crop style: random
Crop aspect: square

anatomy-crop

Repeats: 2
Total number of images: ~16425
Total number of aspect buckets: 1
Resolution: 1.048576 megapixels
Cropped: True
Crop style: random
Crop aspect: square

yoga-crop

Repeats: 0
Total number of images: ~3618
Total number of aspect buckets: 1
Resolution: 1.048576 megapixels
Cropped: True
Crop style: random
Crop aspect: square

photo-aesthetics-crop

Repeats: 0
Total number of images: ~33141
Total number of aspect buckets: 1
Resolution: 1.048576 megapixels
Cropped: True
Crop style: random
Crop aspect: square

text-1mp-crop

Repeats: 125
Total number of images: ~13194
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 = "A photo-realistic image of a cat"

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")

bghira
/

flux-sfwbooru-3.5M-lokr-attempt2