Holy moly! 20k steps! great job! on example pics. I hit 10k steps and my vaginas were still deformed so I just gave up. What were your settings?!

Here is mine that failed

--- job: "extension" config: name: "P0ssy3" process: - type: "diffusion_trainer" training_folder: "D:\\StabilityMatrix-win-x64\\ai-toolkit\\output" sqlite_db_path: "./aitk_db.db" device: "cuda" trigger_word: "p0ssy" performance_log_every: 10 network: type: "lora" linear: 32 linear_alpha: 32 conv: 16 conv_alpha: 16 lokr_full_rank: true lokr_factor: -1 network_kwargs: ignore_if_contains: [] save: dtype: "bf16" save_every: 250 max_step_saves_to_keep: 10 save_format: "diffusers" push_to_hub: false datasets: - folder_path: "D:\\StabilityMatrix-win-x64\\ai-toolkit\\datasets/p0ssy" mask_path: null mask_min_value: 0.1 default_caption: "woman exposing her breasts and vagina " caption_ext: "txt" caption_dropout_rate: 0.05 cache_latents_to_disk: false is_reg: false network_weight: 1 resolution: - 512 - 768 - 1024 controls: [] shrink_video_to_frames: true num_frames: 1 do_i2v: true flip_x: true flip_y: false train: batch_size: 1 bypass_guidance_embedding: false steps: 5000 gradient_accumulation: 1 train_unet: true train_text_encoder: false gradient_checkpointing: true noise_scheduler: "flowmatch" optimizer: "adamw8bit" timestep_type: "weighted" content_or_style: "balanced" optimizer_params: weight_decay: 0.1 unload_text_encoder: false cache_text_embeddings: true lr: 0.0002 ema_config: use_ema: false ema_decay: 0.99 skip_first_sample: true force_first_sample: false disable_sampling: false dtype: "bf16" diff_output_preservation: false diff_output_preservation_multiplier: 1 diff_output_preservation_class: "person" switch_boundary_every: 1 loss_type: "mse" do_differential_guidance: true differential_guidance_scale: 3 model: name_or_path: "Tongyi-MAI/Z-Image-Turbo" quantize: false qtype: "qfloat8" quantize_te: true qtype_te: "qfloat8" arch: "zimage:turbo" low_vram: false model_kwargs: {} layer_offloading: false layer_offloading_text_encoder_percent: 1 layer_offloading_transformer_percent: 1 assistant_lora_path: "ostris/zimage_turbo_training_adapter/zimage_turbo_training_adapter_v1.safetensors" sample: sampler: "flowmatch" sample_every: 250 width: 1024 height: 1024 samples: - prompt: "full body shot of a nude woman with red hair, at the park, sitting down and spreading her legs apart exposing her breasts and vagina" - prompt: "a full body shot of a woman holding a coffee cup, in a beanie, sitting at a cafe with her legs spread wide apart exposing her breasts and vagina" neg: "" seed: 42 walk_seed: true guidance_scale: 1 sample_steps: 8 num_frames: 1 fps: 1 meta: name: "[name]" version: "1.0"

终于有人做这个了

First model and it's great, well done Po! Whens the WF's coming ;)

Why all loras (not just your) dramatically drop down quality of image?

Unnaturally shaved. Where's the hair?

Not bad. 5/10.

Z-Image is definitely struggling in the female anus department... is screws up the pussies too but the anuses are literally non-existent.

Why is the result I achieved much worse compared to the example picture you provided?

It hasn't even been out a full week and we already have full blown nudity LoRAs. LOL

I cant recommend this one, like most of zLoras, the quality drop just kills zImage...still worth just stay in flux or sdxl.

training data is actually just a workflow; not the image set.

Does not work well with character lora's at higher strengths

it works well with the main ZIT model using res2s,3s_beta57, Beta57 generates lots of blotchy noise people like it ass skin tesxture, eventho it's aweful, this lora smoothens that agressive blochiness out, that's why it pairs well with beta57 or anything that creates blotchy noise. with other models you're gnna have to go down to 0.2 0.16, tho, u ar reducing the efectivness. so...weight your options