Another SDXL LoRA for more explicit content, this time it is for dildos and sex toys. This was a difficult concept for SD 1.5 to master, so we will see how XL can handle it.<ul><li>v0.3:<ul><li>Better support for anime and cartoon/render/pixar styles.</li><li>More variety of lingerie and clothing, not always with breasts out.</li><li>Works at 0.8-1.0 strength, not as overtrained.</li></ul></li><li>v0.2:<ul><li>Epoch 93 and 99 are both pretty good, using 0.7 - 0.8 strength.</li></ul></li><li>v0.1:<ul><li>In general, I think epoch 74 has the best results, using about 0.75 strength.</li></ul></li></ul><h2 id="heading-452">Learned Keywords:</h2><ul><li>primary:<ul><li><code>dildo</code> for v0.3</li><li><code>sexl_dildo</code> for v0.1 and v0.2</li></ul></li><li>others:<ul><li>dildo: 1410</li><li>1girl: 1254</li><li>pussy: 1114</li><li>solo: 1099</li><li>realistic: 1015</li><li>sex toy: 864</li><li>object insertion: 837</li><li>nude: 674</li><li>ass: 634</li><li>breasts: 610</li><li>masturbation: 603</li><li>anus: 569</li><li>female masturbation: 563</li><li>nipples: 549</li><li>spread legs: 534</li><li>long hair: 524</li><li>pubic hair: 510</li><li>vaginal: 505</li><li>vaginal object insertion: 488</li><li>female pubic hair: 424</li><li>lips: 360</li><li>lying: 354</li><li>brown hair: 349</li><li>anal: 338</li><li>blonde hair: 317</li><li>black hair: 317</li><li>anal object insertion: 314</li></ul></li></ul>Complete list of tag frequencies: <a target="_blank" rel="ugc" href="https://gist.github.com/onnx-nsfw/4bd77d82a0354853bec14bc667b1ddd1">https://gist.github.com/onnx-nsfw/4bd77d82a0354853bec14bc667b1ddd1</a><h3 id="heading-453">Example Captions:</h3><ul><li>dildo, solo, ass, pantyhose, nude, pussy, indoors, from behind, vaginal, fingernails, uncensored, anus, sex toy, anal, close-up, masturbation, female masturbation, object insertion, realistic, dildo, ass focus, anal object insertion, vaginal object insertion, lower body, gaping, large insertion, dildo riding</li><li>dildo, solo, ass, pantyhose, pussy, from behind, vaginal, uncensored, anus, sex toy, anal, close-up, masturbation, female masturbation, object insertion, realistic, dildo, ass focus, anal object insertion, vaginal object insertion, lower body, gaping, large insertion, dildo riding</li><li>dildo, solo, ass, nude, pussy, from behind, vaginal, fingernails, uncensored, anus, sex toy, anal, close-up, masturbation, female masturbation, object insertion, realistic, dildo, ass focus, anal object insertion, vaginal object insertion, dildo riding</li><li>dildo, solo, thighhighs, underwear, panties, ass, from behind, bra, black panties, back, underwear only, lingerie, black bra, close-up, garter belt, head out of frame, ass focus</li><li>dildo, solo, thighhighs, underwear, panties, ass, from behind, fishnets, sex toy, anal, garter belt, object insertion, dildo, ass focus, fishnet pantyhose, anal object insertion, fishnet thighhighs</li></ul><h2 id="heading-451">Training Method:</h2>Testing this method: <a target="_blank" rel="ugc" href="https://civarchive.com/articles/1771">https://civarchive.com/articles/1771</a>. Versions 0.1 and 0.2 used the 100 epoch method with 2e-05 learning rate.<ul><li>v0.3:<ul><li>1458 source images (&gt; 768 in shortest dimension)</li><li>1588 random crops (1024x1024)</li><li>1410 remaining after curation<ul><li>up to 12 crops per source</li><li>downscaled shortest dimension to 1024</li><li>deduplicated with resnet 18, cosine similarity, 0.925 threshold</li><li>deleted partial and out of frame</li><li>no watermarks</li></ul></li><li>automated captions with WD 1.4, threshold 0.45<ul><li>incorrect tags removed</li></ul></li><li>2 repeats</li><li>8 batch size</li><li>100 epochs</li><li>32 dim/32 alpha</li><li>AdamW8bit with constant</li><li>2e-05 learning rate for both text encoder and UNet</li><li>shuffle captions</li><li>caption dropout rate 0.05</li><li>stop text encoder training after 30% (still not sure if this feature is implemented)</li></ul></li><li>v0.2:<ul><li>344 source images</li><li>287 after deduplication and resize</li><li>1087 random crops</li><li>806 remaining after curation<ul><li>same method as v0.1</li></ul></li><li>automation captions with WD 1.4, threshold 0.6<ul><li>lightly curated, incorrect tags removed</li></ul></li><li>3 repeats</li><li>8 batch size</li><li>100 epochs</li><li>32 dim/32 alpha</li><li>AdamW8bit with constant</li><li>2e-05 learning rate for both text encoder and UNet</li><li>shuffle captions</li></ul></li><li>v0.1:<ul><li>about 500 source images (&gt; 1024x1024)</li><li>about 1000 random crops (1024x1024)</li><li>440 remaining after curation<ul><li>deduplicated with resnet 18, cosine similarity, 0.9 threshold</li><li>deleted partial and out of frame</li><li>no watermarks</li></ul></li><li>1 repeats</li><li>8 batch size</li><li>100 epochs (models around 45-50 and 70-75 were best)</li><li>32 dim/32 alpha</li><li>AdamW8bit with constant</li><li>2e-05 learning rate for both text encoder and UNet</li></ul></li></ul>I am happy to share the data set but do not have a good place for hosting those.<ul><li>v0.3:<ul><li>Includes more lingerie and clothing</li></ul></li><li>v0.2:<ul><li>Next time I need to indicate the color of the dildo in the caption and include more of the image poses.</li></ul></li><li>v0.1:<ul><li>Next time I think I will try removing the <code>sexl_dildo</code> keyword and specializing the existing dildo token.</li></ul></li></ul><h3 id="heading-454">Training Parameters:</h3><pre><code>{
 "LoRA_type": "Standard",
 "adaptive_noise_scale": 0,
 "additional_parameters": "",
 "block_alphas": "",
 "block_dims": "",
 "block_lr_zero_threshold": "",
 "bucket_no_upscale": false,
 "bucket_reso_steps": 64,
 "cache_latents": true,
 "cache_latents_to_disk": true,
 "caption_dropout_every_n_epochs": 0.0,
 "caption_dropout_rate": 0.05,
 "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": false,
 "epoch": 100,
 "factor": -1,
 "flip_aug": true,
 "full_bf16": false,
 "full_fp16": false,
 "gradient_accumulation_steps": "1",
 "gradient_checkpointing": true,
 "keep_tokens": "0",
 "learning_rate": 2e-05,
 "logging_dir": "",
 "lora_network_weights": "",
 "lr_scheduler": "constant",
 "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": "100",
 "max_train_steps": "",
 "mem_eff_attn": false,
 "mid_lr_weight": "",
 "min_bucket_reso": 256,
 "min_snr_gamma": 5,
 "min_timestep": 0,
 "mixed_precision": "fp16",
 "model_list": "custom",
 "module_dropout": 0,
 "multires_noise_discount": 0,
 "multires_noise_iterations": 0,
 "network_alpha": 32,
 "network_dim": 32,
 "network_dropout": 0,
 "no_token_padding": false,
 "noise_offset": 0,
 "noise_offset_type": "Original",
 "num_cpu_threads_per_process": 2,
 "optimizer": "AdamW8bit",
 "optimizer_args": "",
 "output_dir": "/mnt/training/results/dildo-v03-std/",
 "output_name": "last",
 "persistent_data_loader_workers": false,
 "pretrained_model_name_or_path": "stabilityai/stable-diffusion-xl-base-1.0",
 "prior_loss_weight": 1.0,
 "random_crop": false,
 "rank_dropout": 0,
 "reg_data_dir": "",
 "resume": "",
 "sample_every_n_epochs": 1,
 "sample_every_n_steps": 0,
 "sample_prompts": "beautiful woman playing with dildo, detailed, 8k, long hair, breasts, open mouth, multiple girls, ass, nude, pussy, sex toy, anal, object insertion, realistic, dildo, anal object insertion --d 5 --w 1024 --h 1024 --l 9 --s 75",
 "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": "fp16",
 "save_state": 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": "",
 "shuffle_caption": true,
 "stop_text_encoder_training_pct": 30,
 "text_encoder_lr": 2e-05,
 "train_batch_size": 8,
 "train_data_dir": "/mnt/training/cropped/dildo-v03",
 "train_on_input": true,
 "training_comment": "",
 "unet_lr": 2e-05,
 "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"
}</code></pre>
<hr />

Created by <a target="_blank" rel="ugc" href="https://thalis.dev">Thalis AI</a> — dark fantasy, cosmic horror, and the spaces between.

☕ <a target="_blank" rel="ugc" href="https://ko-fi.com/thalisai?utm_source=civitai&amp;utm_medium=model_desc">Support on Ko-fi</a> — every tip keeps the radium gems glowing.
🌙 <a target="_blank" rel="ugc" href="https://www.patreon.com/ThalisAI?utm_source=civitai&amp;utm_medium=model_desc">Join on Patreon</a> — early access, prompt packs, and the deeper corridors.
🔮 <a target="_blank" rel="ugc" href="https://thalis.dev/models/?utm_source=civitai&amp;utm_medium=model_desc">Full catalog</a> — 100+ LoRAs for Flux, Illustrious, and more.

SDXL dildo

v0.3 epoch 97. Includes a greater variety of lingerie and clothing in the training set, works better with anime and 2.5D/3D models.Note: v0.3 no longer uses the <code>sexl_dildo</code> keyword and now uses the regular <code>dildo</code> keyword, which seems to work equally well or better.

Files with SHA256: 6a4fe9b7a4ab31c987b94b56fe489b10c64842dfea57c43a50e3328625100976