Ver NOOB V-PRED EQ V2 GLORA
Trained with GLORA method based on NOOB V-PRED 1.0 with EQ VAE: https://civarchive.com/models/1858821/noobai-v-pred-10-with-eq-vae
See about this version for training details. Make sure your SD software supports GLORA like lateset reforge or Machina's fork when using it.
In this version, I removed half of the concepts that can be directly generated by NOOB to reduce the number of training iterations and avoid overtraining. Please refer to the examples to get a generation like handjob while doggy.
The remained concept dataset was supplemented to balanced ones with AI generated images using the previous version of this lora. Compared to previous versions, it has better generalization ability and stability, and its impact on artstyle has also been reduced. Check the examples for detail.
Ver NOOB V-PRED V1
Trained based on NOOB V-PRED 1.0 with OneTrainer. See about this version for training details. Although NOOB has been able to generate most of the while penetrating scenes, some still cannot be generated and some require a lot of trials. Therefore I think this update is still necessary.
Introduction
A Pony based LyCORIS to generate the concept ** while penetrating. 14 trigger words set. See the examples first.
Pros
Concept generation
Cons
Not responsive to trigger words, you need to upweight it or roll for it times to times
Many trials needed
Strong influence on art style
Some keywords may cause severe limb distortion, which may be caused by overbake
Training details
See about the version.
Description
{
"LoRA_type": "LyCORIS/LoCon",
"LyCORIS_preset": "full",
"adaptive_noise_scale": 0,
"additional_parameters": "",
"async_upload": false,
"block_alphas": "",
"block_dims": "",
"block_lr_zero_threshold": "",
"bucket_no_upscale": true,
"bucket_reso_steps": 128,
"bypass_mode": false,
"cache_latents": false,
"cache_latents_to_disk": false,
"caption_dropout_every_n_epochs": 0,
"caption_dropout_rate": 0,
"caption_extension": ".txt",
"clip_skip": 2,
"color_aug": false,
"constrain": 0,
"conv_alpha": 4,
"conv_block_alphas": "",
"conv_block_dims": "",
"conv_dim": 8,
"dataset_config": "",
"debiased_estimation_loss": false,
"decompose_both": false,
"dim_from_weights": false,
"dora_wd": false,
"down_lr_weight": "",
"dynamo_backend": "no",
"dynamo_mode": "default",
"dynamo_use_dynamic": false,
"dynamo_use_fullgraph": false,
"enable_bucket": true,
"epoch": 10,
"extra_accelerate_launch_args": "",
"factor": -1,
"flip_aug": false,
"fp8_base": false,
"full_bf16": false,
"full_fp16": false,
"gpu_ids": "",
"gradient_accumulation_steps": 1,
"gradient_checkpointing": true,
"huber_c": 0.1,
"huber_schedule": "snr",
"huggingface_path_in_repo": "",
"huggingface_repo_id": "",
"huggingface_repo_type": "",
"huggingface_repo_visibility": "",
"huggingface_token": "",
"ip_noise_gamma": 0.1,
"ip_noise_gamma_random_strength": false,
"keep_tokens": 1,
"learning_rate": 1,
"log_tracker_config": "",
"log_tracker_name": "",
"log_with": "tensorboard",
"logging_dir": "/root/autodl-tmp/output",
"loss_type": "l2",
"lr_scheduler": "cosine",
"lr_scheduler_args": "",
"lr_scheduler_num_cycles": 1,
"lr_scheduler_power": 1,
"lr_warmup": 0,
"main_process_port": 0,
"masked_loss": false,
"max_bucket_reso": 4096,
"max_data_loader_n_workers": 0,
"max_grad_norm": 1,
"max_resolution": "1024,1024",
"max_timestep": 1000,
"max_token_length": 75,
"max_train_epochs": 0,
"max_train_steps": 0,
"mem_eff_attn": false,
"metadata_author": "",
"metadata_description": "",
"metadata_license": "",
"metadata_tags": "",
"metadata_title": "",
"mid_lr_weight": "",
"min_bucket_reso": 1024,
"min_snr_gamma": 0,
"min_timestep": 0,
"mixed_precision": "bf16",
"model_list": "custom",
"module_dropout": 0,
"multi_gpu": false,
"multires_noise_discount": 0.1,
"multires_noise_iterations": 6,
"network_alpha": 8,
"network_dim": 16,
"network_dropout": 0,
"network_weights": "",
"noise_offset": 0,
"noise_offset_random_strength": false,
"noise_offset_type": "Multires",
"num_cpu_threads_per_process": 2,
"num_machines": 1,
"num_processes": 1,
"optimizer": "Prodigy",
"optimizer_args": "\"weight_decay=0.01\" \"d_coef=1\" \"use_bias_correction=True\" \"safeguard_warmup=False\" \"betas=0.9,0.99\" \"decouple=True\"",
"output_dir": "/root/autodl-tmp/output",
"output_name": "While Penetrating-PonyXL",
"persistent_data_loader_workers": false,
"pretrained_model_name_or_path": "/root/kohya_ss/models/ponyDiffusionV6XL.safetensors",
"prior_loss_weight": 1,
"random_crop": false,
"rank_dropout": 0,
"rank_dropout_scale": false,
"reg_data_dir": "/root/autodl-tmp/reg_while_penetrating",
"rescaled": false,
"resume": "",
"resume_from_huggingface": "",
"sample_every_n_epochs": 1,
"sample_every_n_steps": 0,
"sample_prompts": "score_9, score_8_up, score_7_up, anime BREAK, handjob_doggy, long hair, breasts, blush, large breasts, multiple girls, 2girls, nipples, closed eyes, pink hair, nude, penis, tongue, pointy ears, cum, orange hair, huge breasts, collar, uncensored, bed, grabbing, grabbing another's breast, kiss, sex toy, handjob, futanari, futa with female, french kiss, artificial vagina, futa with futa, futasub --n source_pony, source_furry, source_cartoon, greyscale, from below --w 1024 --h 1024\n ",
"sample_sampler": "euler_a",
"save_every_n_epochs": 1,
"save_every_n_steps": 0,
"save_last_n_steps": 0,
"save_last_n_steps_state": 0,
"save_model_as": "safetensors",
"save_precision": "bf16",
"save_state": false,
"save_state_on_train_end": false,
"save_state_to_huggingface": false,
"scale_v_pred_loss_like_noise_pred": false,
"scale_weight_norms": 0,
"sdxl": true,
"sdxl_cache_text_encoder_outputs": false,
"sdxl_no_half_vae": true,
"seed": 31337,
"shuffle_caption": true,
"stop_text_encoder_training_pct": 0,
"text_encoder_lr": 1,
"train_batch_size": 5,
"train_data_dir": "/root/autodl-tmp/while_penetrating",
"train_norm": false,
"train_on_input": true,
"training_comment": "",
"unet_lr": 1,
"unit": 1,
"up_lr_weight": "",
"use_cp": false,
"use_scalar": false,
"use_tucker": false,
"v2": false,
"v_parameterization": false,
"v_pred_like_loss": 0,
"vae": "/root/kohya_ss/models/sdxl_vae_fp16fix.safetensors",
"vae_batch_size": 0,
"wandb_api_key": "",
"wandb_run_name": "",
"weighted_captions": false,
"xformers": "xformers"
}FAQ
Comments (6)
this is good and all, but i now wonder if you can make a lora based on two to three futa girls just masturbating each other, no penetration.
That's a good idea. I will try.
What I like most about this, is not how great the results are - which they are; but how much this model has taught me about how weight affects things like limb generation. For months I used to throw some numbers at the wall and just went with what stuck, but now I finally realize that even a slight difference of 0.05 weight can mean the difference between a beautifully generated image and freaky limb spaghetti.
Thank you so much, this has been better than every SD tutorial ever.
The strategy to make this lora consistent but also not compromise the artstyle is to use Lora Block Weights with something like INALL and set both the Lora weight and keyword weight to around 1.4.
Please Illustrious version of this masterpiece
Thank you for your feedback. But you may try some similar prompt with NOOB. I tried to train it on NOOB V0.5 once and found it achieve similar generations without this Lora. See my post in NOOB to get an example.