thick thighs

sexual situations

physical violence

disturbing

extremist

hanging

hate symbols

nazi party

breasts out

strapless leotard

revealing clothes

weapon violence

female swimwear or underwear

male swimwear or underwear

partial nudity

self injury

white supremacy

adult toys

graphic male nudity

illustrated explicit nudity

nudity

sexual activity

graphic violence or gore

graphic female nudity

vore

bestiality

corpses

breast out

one breast out

huge breasts

gigantic breasts

huge butt

wide hips

covered nipples

hair over breasts

no panties

sitting on face

convenient censoring

anal

dildo riding

downblouse

peeing

oral

oral invitation

emaciated bodies

hentai

porn

bukkake

cumshot

cumdrip

cum in pussy

cum on face

cum on hair

cum on body

cum on tongue

cum on hands

cum in mouth

cum on pussy

eat_cum

fellatio

implied fellatio

after fellatio

triple fellatio

autofellatio

pov fellatio

fucked silly

presenting ass

presenting pussy

spreading legs

spread legs

swimwear

lingerie

bikini

latex clothing

TRPCError

Flux version added for that late 80's early 90's anime aesthetic Trained on Satoshi Urushihara artwork and OVA stills.

resized with --new_rank 64 --save_precision bf16 --dynamic_method sv_fro --dynamic_param 0.96

Flux v3 Resize

experimental Flux dev version, 1286 images, joycaptioned, command line:<pre><code>accelerate launch --mixed_precision bf16 --num_cpu_threads_per_process 1 flux_train_network.py --pretrained_model_name_or_path /workspace/train/flux1-dev.sft --clip_l /workspace/train/clip_l.safetensors --t5xxl /workspace/train/t5xxl_fp16.safetensors --ae /workspace/train/ae.sft --cache_latents_to_disk --save_model_as safetensors --sdpa --persistent_data_loader_workers --max_data_loader_n_workers 2 --seed 42 --gradient_checkpointing --mixed_precision bf16 --save_precision bf16 --network_module networks.lora_flux --network_dim 64 --optimizer_type adamw8bit --learning_rate 0.00005 --network_train_unet_only --cache_text_encoder_outputs --cache_text_encoder_outputs_to_disk --highvram --max_train_epochs 16 --save_every_n_epochs 1 --dataset_config /workspace/train/urushihara.toml --output_dir /workspace/train --output_name urushihara3 --timestep_sampling shift --model_prediction_type raw --guidance_scale 1.0 --loss_type l2</code></pre>

Flux v3

XL version trained on the same dataset as the 1.5 model --pretrained_model_name_or_path="/workspace/train/sd_xl_base_1.0.safetensors" --vae="/workspace/train/sdxl_vae.safetensors" --caption_extension=".txt" --train_data_dir="/workspace/train/urushihara_satoshi" --output_name="xl_urushihara_satoshi_v16" --resolution="1024,1024" --output_dir="/workspace/train" --logging_dir="/workspace/logs" --save_model_as=safetensors --network_module=networks.lora --save_every_n_epochs="8" --network_dim=128 --network_alpha=64 --train_batch_size="4" --optimizer_type="Prodigy" --optimizer_args "betas=0.9,0.999" "decouple=True" "weight_decay=0.0001" "d0=1e-4" "d_coef=2" "use_bias_correction=True" "safeguard_warmup=True" --learning_rate="1.0" --text_encoder_lr=1.0 --unet_lr=1.0 --max_data_loader_n_workers="0" --max_token_length=225 --keep_tokens="1" --enable_bucket --bucket_reso_steps=64 --min_bucket_reso=512 --max_bucket_reso=2048 --max_train_epochs="32" --network_train_unet_only --random_crop --no_half_vae --mixed_precision="fp16" --save_precision="fp16" --xformers --lr_scheduler_type "CosineAnnealingLR" --lr_scheduler_args "T_max=10496"

XL v16

this is a huge 384 dim LoRA, experimenting with adaptive learning rate note I tried resizing this LoRA to lower dims using the kohya_ss tools, but the quality results were too subjective (some images degrade, others are improved) training settings<ul><li>3137 training images</li><li>512,512 training resolution</li><li>16 epochs, 6,384 steps, 8 batches</li><li>network dim &amp; alpha 384</li><li>DAdaptAdam optimizer, constant scheduler</li><li>learning rate 1.0</li><li>bucketing &amp; random crop</li></ul>useful optional tokens<ul><li>urushisato tagged on all images for general style boost</li><li>ova tagged on all screencaps from animation</li><li>background tagged on any image that has no human subject</li></ul>

urushisato_v15

urushisato_v7

Satoshi Urushihara style

Type	LoRA
Stats	623
Reviews	Very Positive (61)
Published	Aug 31, 2024
Base Model	Flux.1 D
Training	Epochs: 16
Usage Tips	Clip Skip: 1
Hash	AutoV2 6CAE3743FB