New version includes lots more high res training material. Almost everything is trained at 832x1216 or 1216x832
Uncut has been spun off into its own model https://civarchive.com/models/418871/xpenis-uncut
Some keywords: flaccid penis, erect penis, small flaccid penis, big flaccid penis, small erect penis, big erect penis.
You have to use a VAE, no VAE is baked in. Works best with SDXL base or low CFG for other models
Description
New version trained on higher quality, and more diverse curated material. I use comfy-ui.
My sample images are all single shot txt2img, no img2img, no inpainting, no upscale, no controlnet.
1024x1024 at 30 steps
FAQ
Comments (11)
Thank you for your work! I could get some nice results using JuggernautXL I hope you can keep updating this!
Thanks for the effort but it affects the checkpoints drastically. They look all samey if you generate a few of them. They dont look unique after some time.
And If you reduce the lora weight often no penis is rendered.
I can get only okayish results with a lot of prompting and with the help of Controlnet to force the penis to be rendered.
Please keep up the work!
Maybe there is a way to less affect the checkpoints and other loras.
How to you use it? I use it like shown in the comfy screenshot and it seems very capable of making diverse people from twinks to old guys, fat, muscle guys, various races etc. Check out the samples.
Did you put penis as a trigger word?
As it tends towards muscle guys but if you don't want that add muscle to your negative prompt or add fat or belly to you positive.
The checkpoints i used it with were crystal and juggernaut.
My setting was around 0.6 weight.
@wolffur666456 I am using it as Lora in webUI. I am using penis keywords etc. The lora is affecting the surroundings a lot and changing the character, It´s often getting mushy and the people looks very similar.
ALso have problems when combining with other Loras.
I tried all settings...when getting higher than 0.5 it changed the look of the character too much.
Used
CopaxTimelessSDXL
JuggernautXL
Rundiffusion
Realvisxl
@CyclopsGER I usually keep it around 0.5. Nothing else is anywhere close to as good as this right now.
@benjimon678 You are right there and I am thankful for this! But I think feedback is important.
@CyclopsGER I highly suggest trying comfyui. It's also just faster.
As for messing with the other loras. I wonder if they were trained on different VAEs. I used the default SDXl base checkpoint and VAE for training. Can you give me an example of one of the loras you are testing with so I can experiment with it
there is no real way around it without using checkpoints that understand what a penis is
@wolffur666456 I never actually tried this with the base model. I will also check the other loras first without a penis. Maybe they are causing the issues.
I am working on an anal sex lora. I made a model merge with xpenis and have been testing alot of things. Got a few good results. Could you share your learning rates, optimizer and image set size used to create this lora?
{
"LoRA_type": "Standard",
"adaptive_noise_scale": 0,
"additional_parameters": "",
"block_alphas": "",
"block_dims": "",
"block_lr_zero_threshold": "",
"bucket_no_upscale": true,
"bucket_reso_steps": 64,
"cache_latents": true,
"cache_latents_to_disk": true,
"caption_dropout_every_n_epochs": 0.0,
"caption_dropout_rate": 0,
"caption_extension": "txt",
"clip_skip": "1",
"color_aug": false,
"conv_alpha": 1,
"conv_block_alphas": "",
"conv_block_dims": "",
"conv_dim": 1,
"decompose_both": false,
"dim_from_weights": false,
"down_lr_weight": "",
"enable_bucket": true,
"epoch": 30,
"factor": -1,
"flip_aug": false,
"full_bf16": false,
"full_fp16": false,
"gradient_accumulation_steps": "1",
"gradient_checkpointing": false,
"keep_tokens": "0",
"learning_rate": 0.00015,
"logging_dir": "/dataset/sdxl/xpenis/published/1.4/log",
"lora_network_weights": "",
"lr_scheduler": "constant_with_warmup",
"lr_scheduler_args": "",
"lr_scheduler_num_cycles": "",
"lr_scheduler_power": "",
"lr_warmup": 0,
"max_bucket_reso": 2048,
"max_data_loader_n_workers": "0",
"max_resolution": "1024,1024",
"max_timestep": 1000,
"max_token_length": "75",
"max_train_epochs": "",
"max_train_steps": "",
"mem_eff_attn": false,
"mid_lr_weight": "",
"min_bucket_reso": 256,
"min_snr_gamma": 0,
"min_timestep": 0,
"mixed_precision": "bf16",
"model_list": "custom",
"module_dropout": 0,
"multires_noise_discount": 0,
"multires_noise_iterations": 0,
"network_alpha": 64,
"network_dim": 128,
"network_dropout": 0,
"no_token_padding": false,
"noise_offset": 0,
"noise_offset_type": "Original",
"num_cpu_threads_per_process": 16,
"optimizer": "Adafactor",
"optimizer_args": "scale_parameter=False relative_step=False warmup_init=False",
"output_dir": "/dataset/sdxl/xpenis/published/1.4/model",
"output_name": "xpenis-1.4",
"persistent_data_loader_workers": false,
"pretrained_model_name_or_path": "/training-models/StableDiffusion/_regular/xl/sd_xl_base_1.0.safetensors",
"prior_loss_weight": 1.0,
"random_crop": false,
"rank_dropout": 0,
"reg_data_dir": "",
"resume": "",
"sample_every_n_epochs": 0,
"sample_every_n_steps": 50,
"sample_prompts": "naked man with big penis and testicles, standing in an elevator, embarassed, hairy chest, smiling, handsome, beard, view from below --w 1024 --h 1024 --s 28 --d 87289379384 ",
"sample_sampler": "euler_a",
"save_every_n_epochs": 5,
"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": true,
"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": "",
"shuffle_caption": false,
"stop_text_encoder_training": 0,
"text_encoder_lr": 0.0,
"train_batch_size": 1,
"train_data_dir": "/dataset/sdxl/xpenis/published/1.4/img",
"train_on_input": true,
"training_comment": "",
"unet_lr": 0.0,
"unit": 1,
"up_lr_weight": "",
"use_cp": false,
"use_wandb": false,
"v2": false,
"v_parameterization": false,
"v_pred_like_loss": 0,
"vae_batch_size": 0,
"wandb_api_key": "",
"weighted_captions": false,
"xformers": "xformers"
}