configs/NOOB_BASE.knf.yaml

# ==============================================================
# == Directories
# ==============================================================


# > Dataset sync derives SOURCE/, MANIFEST.knf.xlsx, and HF/<config-id>/ from this filename.

# -- Map exported subset names to source subfolders in MANIFEST.knf.xlsx.
mappings:
  # > Anchor identity examples get sampled at a lower probability below.
  felkin_anchor:
    - "0-ANCHOR" # Source folder for anchor identity images.
  # > Explicit anatomy close-up subsets are kept separate for sampling control.
  genitals:
    - "6-DICK"   # Source folder for exposed anatomy close-ups.
    - "6-SHEATH" # Source folder for sheathed anatomy close-ups.
  # > Complete body.
  fullbody:
    - "0-FULLBODY" # Full-body identity examples; also included in rest subset.
    - "99-VERSION-LEANER" # Alternate leaner full-body examples.
  # > General body, hand, leg, tail, and leaner variant images.
  body_details:
    - "1-TORSO" # Torso detail examples.
    - "3-HANDS" # Hand detail examples.
    - "4-LEGS" # Leg/foot detail examples.
    - "5-TAIL" # Tail detail examples.
  # > Focused fur-texture crops derived from high-resolution Rook sources.
  fur:
    - "7-FUR" # Fur crops for coat, mane, spots, paws, tail, and anatomy-adjacent texture.
  head:
    - "2-HEAD" # Face/head detail examples.

# > Also export manifest rows marked sfw into an additional generated sfw subset.
export_sfw_subset: false # < Keep NOOB focused on the explicit mapped subsets.

# ==============================================================
# == Images
# ==============================================================

# Export-time mirroring writes fixed extra image/caption pairs before
# SimpleTuner sees the dataset. flip_only keeps kneifftools from applying its
# own random crop, rotation, or color jitter.
augmentations:
  mirrored_extra: true           # Add one mirrored copy per selected source image.
  mirrored_transform: flip_only  # Exact horizontal mirror, no kneifftools crop/jitter.
  seed: 12345                    # Makes mirrored export naming deterministic.

# Export image area bounds before SimpleTuner sees the dataset.
image_resize:
  min_pixel_area: 512   # Upscale tiny images below roughly 512x512 area.
  max_pixel_area: 1024  # Downscale large images above roughly 1024x1024 area.

# ==============================================================
# == Captions
# ==============================================================

# Controlled caption rendering context used by all generated sidecars.
caption:
  subject_sex: male           # Adds male subject wording to prose renderers.
  add_domains_to_tags: false  # SDXL tag captions should stay plain, not domain-prefixed.

# NOOB uses SDXL-style tag captions. Use one .txt sidecar per image so
# SimpleTuner's textfile strategy reads the intended caption directly.
caption_outputs:
  mode: hybrid_txt # One .txt sidecar per image containing selected formats.
  formats: [tag]   # Tag-only captions for NOOB/SDXL training.
  tag_scope: all   # Keep the full tag tail, including identity and anatomy tags.

# Optional per-subset caption output overrides. Keep commented unless a subset
# needs different caption files or formats.
# caption_outputs_overrides:
#   felkin_anchor:
#     mode: hybrid_txt
#     formats: [tag]
#     tag_scope: all
#   genitals:
#     mode: hybrid_txt
#     formats: [tag]
#     tag_scope: all
#   fullbody:
#     mode: hybrid_txt
#     formats: [tag]
#     tag_scope: all
#   body_details:
#     mode: hybrid_txt
#     formats: [tag]
#     tag_scope: all
#   fur:
#     mode: hybrid_txt
#     formats: [tag]
#     tag_scope: all
#   head:
#     mode: hybrid_txt
#     formats: [tag]
#     tag_scope: all
#   sfw:
#     mode: hybrid_txt
#     formats: [tag]
#     tag_scope: all

# ==============================================================
# == Publishing
# ==============================================================

publishing:
  huggingface:
    enabled: true # Generate dataset-card metadata during dataset sync.
    repo_id: null # Optional Hugging Face dataset repo id.
    pretty_name: Rook NOOB SDXL LoRA # Human-readable dataset-card base title.
    version: null # Public dataset version label.
    optimized_for_model: Willys Noob Realism Core V1 # Rendered as [For Willys Noob Realism Core V1].
    license: null # Optional dataset license.
    tags: [sdxl, noob, lora] # Dataset-card tags.
    provenance: null # Optional dataset provenance note.
    adult_content: false # Render the adult-content notice in README.md.
    notes: null # Optional free-form dataset-card notes.

# ==============================================================
# == LoRA Training
# ==============================================================

training:
  enabled: true        # Enables kneifftools training artifact generation.
  backend: simpletuner # Selects the SimpleTuner LoRA backend.

  simpletuner:
    enabled: true # Optional backend-specific switch; defaults to enabled.

    # Subset-specific SimpleTuner dataloader overrides. Any key here is merged
    # into the subset backend entry after the dataset defaults below.
    subsets:
      felkin_anchor:
        probability: 0.01          # Sample anchor identity images at steady exposure.
        crop: false             # Preserve full source composition.
        # repeats: 0            # Optional per-subset repeat count.
        # disabled: false       # Optional per-subset on/off switch.
      fullbody:
        probability: 10          # Sample full-body subset at high probability.
        crop: false             # Preserve full-body composition.
        # repeats: 0
        # disabled: false
      body_details:
        probability: 4          # Sample body detail subset at normal probability.
        crop: false             # Preserve detail framing; many samples are already close crops.
        # repeats: 0
        # disabled: false
      fur:
        probability: 1          # Give focused fur crops extra exposure.
        crop: false             # Preserve texture crop intent; do not random-crop fur samples.
        # repeats: 0
        # disabled: false
      head:
        probability: 3          # Sample head/face details at extra exposure.
        crop: false             # Preserve detail framing; many samples are close crops.
      genitals:
        probability: 3          # Sample anatomy subset at normal probability.
        crop: false             # Preserve anatomy crop framing.
        # repeats: 0
        # disabled: false

    # Dataset backend defaults copied into every image subset entry.
    dataset:
      type: local                  # SimpleTuner local filesystem backend.
      dataset_type: image          # Image dataset backend, not text embeds.
      resolution: 1024             # Target area edge; pixel_area means 1024^2 area.
      minimum_image_size: 512      # SimpleTuner lower bound for SDXL image handling.
      maximum_image_size: 1024     # Upper bound for input image sizing.
      target_downsample_size: 1024 # Downsample target for cache generation.
      resolution_type: pixel_area  # Area-based buckets, not smaller-edge resize.
      caption_strategy: textfile   # Read captions from exported .txt sidecars.
      metadata_backend: discovery  # Discover images/captions from directories.
      crop: false                  # Default only; every configured subset disables crop above.
      crop_style: random           # Inert for crop:false subsets.
      crop_aspect: preserve        # Preserve original aspect instead of square crop.
      repeats: 0                   # Dataset-level repeat count; subset can override.
      disabled: false              # Dataset-level backend switch.

    # Optional shared text-embedding backend overrides. Commented keys show the
    # kneifftools-owned defaults for the generated text_embeds backend.
    text_embeds:
      id: text-embeds       # Backend id used by SimpleTuner.
      type: local           # Store embeddings on local filesystem.
      dataset_type: text_embeds
      default: true         # Marks this as the shared embedding backend.
      # cache_dir: .simpletuner-cache/text # Override generated text cache dir.
      disabled: false

    # Generate a SimpleTuner user_prompt_library JSON from manifest-sampled tag
    # prompts plus a few stable NOOB probes.
    validation_prompts:
      enabled: true
      positive_prefix: "masterpiece, best quality, newest, "
      from_manifest:
        profile: tag
        seed: 42
      custom:
        noob_safe_portrait: "Rook_Kaefer, felkin, fur, red fur, black spots, spotted fur, fur markings, close-up portrait, detailed lighting"
        noob_safe_fullbody_reference: "Rook_Kaefer, felkin, fur, red fur, black spots, spotted fur, fur markings, musclegut, full-body character reference, standing, solo"
        noob_safe_fur_detail: "Rook_Kaefer, felkin, fur, red fur, black spots, spotted fur, fur markings, musclegut, fur texture reference, mane, tail, paws"
        noob_explicit_anatomy_reference: "Rook_Kaefer, felkin, fur, red fur, black spots, spotted fur, fur markings, musclegut, nude anatomy reference, sheath detail, clean lighting"

    # Model and LoRA settings merged into the top-level SimpleTuner config.
    model:
      model_type: lora       # Train a LoRA adapter.
      model_family: sdxl     # Use SimpleTuner's SDXL model family.
      model_flavour: null    # Loaded from a local SDXL checkpoint.
      lora_rank: 64          # LoRA rank/capacity.
      lora_alpha: 64         # LoRA alpha scaling.
      lora_type: standard    # Standard LoRA, not an alternate adapter type.
      lora_format: diffusers # SDXL LoRA format expected by SimpleTuner for this run.
      flux_lora_target: null # Remove the Flux/Chroma default from generated config.

      # 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
      pretrained_t5_model_name_or_path: null
      pretrained_text_encoder_model_name_or_path: null

      # Precision defaults for SDXL LoRA training.
      base_model_precision: no_change
      base_model_default_dtype: bf16
      mixed_precision: bf16
      # init_lora: /path/to/existing.safetensors # Optional LoRA resume/init weights.
      # lycoris_config: /path/to/lycoris.json    # Optional LyCORIS config path.

    # == Curriculum
    curriculum:
      enabled: true
      phases:
        - name: focused_start
          start_step: 0
          subsets: [fullbody, head]
        - name: full_mix
          start_step: 1000 # !! From this step onward, all subsets are active.
          subsets: all

    # Full-run trainer config. trainer_testrun below overrides selected keys
    # only when knf train prepare/start is called with --testrun.
    trainer:
      # tracker_project_name: kneifftools-lora-training # Experiment tracker project.
      # tracker_run_name: ladybird-noob                 # Full-run tracker name.
      # hub_model_id: ladybird-noob-lora                # Optional Hub model id if push_to_hub is enabled.

      optimizer: adamw_bf16              # BF16 AdamW optimizer.
      learning_rate: 0.00015              # SDXL LoRA learning rates fall between 1e-4 and 5e-4
      lr_scheduler: constant_with_warmup # Ramp up LR to avoid weird color biases.
      lr_warmup_steps: 200
      seed: 42                           # Training seed.
      gradient_checkpointing: true       # Reduce VRAM by recomputing activations.
      # max_grad_norm: 1.0               # Optional gradient clipping.
      # grad_clip_method: norm           # Optional gradient clipping method.

      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: 3200          # Current short SDXL run length.

      caption_dropout_probability: 0.05 # Caption dropout for regularization.
      data_backend_sampling: uniform   # Uniform dataloader sampling.
      allow_dataset_oversubscription: true

      push_to_hub: false             # Upload final artifacts to Hugging Face Hub.
      push_checkpoints_to_hub: false # Upload intermediate checkpoints.
      checkpoints_total_limit: 3   # Keep latest 5 checkpoints.
      checkpoint_step_interval: 400    # Save checkpoint every 200 steps.
      validation_step_interval: 400    # Run validation every 200 steps.
      # disable_benchmark: false  # Render baseline validation before training.
      validation_prompt_library: false       # Disable SimpleTuner built-in prompt library.
      validation_disable_unconditional: true # Skip unconditional validation branch.
      validation_resolution: 1024x1024       # Validation image size.
      validation_num_inference_steps: 50     # Full-run validation sampling steps.
      validation_guidance: 5.0               # SDXL CFG guidance used by validation.
      validation_guidance_real: null         # Remove flow-model real CFG from generated config.
      validation_guidance_rescale: 0.0       # Guidance rescale; 0 disables rescale.
      validation_seed: 42                    # Deterministic validation seed.
      num_eval_images: 1                     # Images per validation prompt.
      # validation_negative_prompt: "signature, ugly, cropped, blurry, low-quality"

      report_to: none # Set to tensorboard to write TensorBoard logs.
      # logging_dir: logs # TensorBoard log dir relative to TRAINING/<config-id>_<run>.

      # Additional kneifftools-owned SimpleTuner defaults. Leave commented to
      # use defaults from src/kneiff/training/lora/simpletuner.py.
      use_ema: false
      vae_batch_size: 1
      aspect_bucket_alignment: 64
      tokenizer_max_length: 77
      t5_padding: null
      fully_unload_text_encoder: true
      offload_during_startup: true
      keep_vae_loaded: false
      resume_from_checkpoint: null

    # Short smoke-run overrides. These merge last only with --testrun.
    trainer_testrun:
      max_train_steps: 4                      # Tiny run to verify plumbing.
      checkpoint_step_interval: 100           # Avoid extra smoke checkpoints.
      validation_step_interval: 4             # Validate at the end of the smoke run.
      ignore_final_epochs: true               # Fix early stop due to SimpleTuner epoch calculation.