- added three-quarter view image.

MANIFEST.knf.xlsx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:12c31e06c34e18a2c39638234ba06db46683ef40411f56a9b081ca8816ed90e0
-size 14809992
+oid sha256:8a667b37d520f359fb55416fa21512775840e4fd0027dbbf77a7cf86693f16fb
+size 14896912

MANIFEST.yaml

@@ -2673,3 +2673,8 @@
   style: photorealistic
   penis_erection: fully_sheathed
   background: simple_background
+- Relative_path: 0-FULLBODY/Three-quarterview/iden_00006_.png
+  Subfolders: Three-quarterview
+  Filename: iden_00006_.png
+  Sync_timestamp: "2026-06-24 23:07:10"
+  Resolution: 1024x1024

configs/NOOB_BASE.knf.yaml

@@ -222,8 +222,9 @@ training:
       lora_format: diffusers # SDXL LoRA format expected by SimpleTuner for this run.
       flux_lora_target: null # Remove the Flux/Chroma default from generated config.
 
-      # Converted full SDXL Diffusers directory from the WSL ComfyUI model store.
-      pretrained_model_name_or_path: /mnt/wsl/comfyui-models/models/checkpoints/_base_finetunes/willysRealism_coreV1Diffusers
+      # Train against the clean NOOB SDXL Diffusers base for adapter compatibility.
+      # Willy stays the intended downstream inference/showcase base after export.
+      pretrained_model_name_or_path: /mnt/wsl/comfyui-models/models/checkpoints/_base_finetunes/noobaiXLNAIXL_epsilonPred11Version
       pretrained_unet_model_name_or_path: null
       pretrained_vae_model_name_or_path: madebyollin/sdxl-vae-fp16-fix
       pretrained_transformer_model_name_or_path: null
@@ -262,7 +263,7 @@ training:
       train_batch_size: 4            # Per-device batch size.
       gradient_accumulation_steps: 1 # Effective batch multiplier.
       num_train_epochs: 0            # Step-based training; max_train_steps controls run length.
-      max_train_steps: 1500          # Current short SDXL run length.
+      max_train_steps: 2400          # Current short SDXL run length.
 
       caption_dropout_probability: 0.0 # Caption dropout for regularization.
       data_backend_sampling: uniform   # Uniform dataloader sampling.