Text-to-Image
Diffusers
template:sd-lora
flux
lora
001_Dark_Industrial / config.yaml
rvorias's picture
Upload folder using huggingface_hub
6f53ed2 verified
config:
name: Dark_Industrial
process:
- datasets:
- cache_latents_to_disk: true
caption_dropout_rate: 0.2
caption_ext: txt
folder_path: /root/lorahub/Dark_Industrial/dataset
resolution:
- 512
- 768
- 1024
shuffle_tokens: false
token_dropout_rate: 0.01
device: cuda:0
model:
is_flux: true
name_or_path: black-forest-labs/FLUX.1-dev
quantize: true
text_encoder_bits: 8
network:
linear: 64
linear_alpha: 64
transformer_only: true
type: lora
performance_log_every: 500
sample:
guidance_scale: 3.5
height: 896
neg: ''
prompts:
- '[trigger]. A giant dark underground industrial spacious rounded location
composed of concrete and rusted steel, filled with water with a smooth reflective
surface and a slight haze.'
- '[trigger]. In this dark location, corridors, walls, openings, floor and ceiling
are built only of thick rusty expanded metal mesh and rusty steel angles.
Behind expanded metal meshes a multilayer structure of the location is visible.
Through the openings, rusty columns and the meshes of another floor are visible.
Rare dim industrial lamps could be seen on the walls.'
- '[trigger]. Living room with sofa and TV, but is underground and has a industrial
feel. Bent rusty steel iron, peeling crumbling concrete walls, industrial
trash, darkness, very dim industrial lamp, bent rebar is embedded in the walls.
Photo.'
- '[trigger], IMG_1025.HEIC RAW photograph: sewer tunnel with water, rebar sticking
out of the walls.'
- '[trigger]. Photo of a statue of a man made only of bent rusty steel rebar.
It''s dark.'
- New Year's tree in an empty apartment, garlands, photo.
sample_every: 500
sample_steps: 25
sampler: flowmatch
seed: 1000
walk_seed: true
width: 1152
save:
dtype: float16
max_step_saves_to_keep: 3
save_every: 500
save_format: diffusers
train:
batch_size: 2
dtype: bf16
ema_config:
ema_decay: 0.99
use_ema: true
gradient_accumulation_steps: 1
gradient_checkpointing: true
linear_timesteps: true
loss_type: mse
lr: 0.0003
noise_scheduler: flowmatch
optimizer: adamw8bit
reg_weight: 1.0
steps: 3000
target_noise_multiplier: 1.0
train_text_encoder: false
train_unet: true
training_folder: /root/lorahub
trigger_word: IndustrialC0re
type: sd_trainer
job: extension
meta:
description: v1