shootthesound
0f1d5a243d
LoRA Analysis & Selective Loading New features for understanding and fine-tuning LoRA block-level behavior: New Nodes: - LoRA Loader + Analyzer - Drop-in replacement for standard LoRA loader that analyzes which blocks have the most impact. Outputs analysis JSON and lora_path for chaining to selective loaders. - Selective LoRA Loader (SDXL/Z-Image/FLUX/Wan/Qwen) - 5 architecture-specific loaders with per-block toggles and strength sliders. Includes presets (Default, All Off, Half Strength, High Impact, Face Focus, etc.) Features: - Impact-colored checkboxes: Connect analyzer's analysis_json output to see blocks color-coded from blue (low impact) to red (high impact) - Path passthrough: Connect lora_path from analyzer or any trainer directly to selective loaders - dropdown is ignored when path is connected - JavaScript preset system for quick block configuration
60 lines
2.8 KiB
Python
60 lines
2.8 KiB
Python
"""
|
|
ComfyUI Realtime LoRA Trainer
|
|
|
|
Trains LoRAs on-the-fly from images during generation.
|
|
Supports Z-Image, FLUX, Wan models via AI-Toolkit.
|
|
Also supports SDXL and SD 1.5 via kohya sd-scripts.
|
|
|
|
Includes LoRA Layer Analyzer and Selective LoRA Loader for analyzing
|
|
and loading specific blocks/layers from LoRA files.
|
|
"""
|
|
|
|
from .realtime_lora_trainer import RealtimeLoraTrainer, ApplyTrainedLora
|
|
from .sdxl_lora_trainer import SDXLLoraTrainer
|
|
from .sd15_lora_trainer import SD15LoraTrainer
|
|
from .musubi_zimage_lora_trainer import MusubiZImageLoraTrainer
|
|
from .musubi_qwen_image_lora_trainer import MusubiQwenImageLoraTrainer
|
|
from .musubi_qwen_image_edit_lora_trainer import MusubiQwenImageEditLoraTrainer
|
|
from .musubi_wan_lora_trainer import MusubiWanLoraTrainer
|
|
from .lora_analyzer import LoRALoaderWithAnalysis
|
|
from .selective_lora_loader import SDXLSelectiveLoRALoader, ZImageSelectiveLoRALoader, FLUXSelectiveLoRALoader, WanSelectiveLoRALoader, QwenSelectiveLoRALoader
|
|
|
|
# Web directory for JavaScript extensions
|
|
WEB_DIRECTORY = "./web/js"
|
|
|
|
NODE_CLASS_MAPPINGS = {
|
|
"RealtimeLoraTrainer": RealtimeLoraTrainer,
|
|
"ApplyTrainedLora": ApplyTrainedLora,
|
|
"SDXLLoraTrainer": SDXLLoraTrainer,
|
|
"SD15LoraTrainer": SD15LoraTrainer,
|
|
"MusubiZImageLoraTrainer": MusubiZImageLoraTrainer,
|
|
"MusubiQwenImageLoraTrainer": MusubiQwenImageLoraTrainer,
|
|
"MusubiQwenImageEditLoraTrainer": MusubiQwenImageEditLoraTrainer,
|
|
"MusubiWanLoraTrainer": MusubiWanLoraTrainer,
|
|
"LoRALoaderWithAnalysis": LoRALoaderWithAnalysis,
|
|
"SDXLSelectiveLoRALoader": SDXLSelectiveLoRALoader,
|
|
"ZImageSelectiveLoRALoader": ZImageSelectiveLoRALoader,
|
|
"FLUXSelectiveLoRALoader": FLUXSelectiveLoRALoader,
|
|
"WanSelectiveLoRALoader": WanSelectiveLoRALoader,
|
|
"QwenSelectiveLoRALoader": QwenSelectiveLoRALoader,
|
|
}
|
|
|
|
NODE_DISPLAY_NAME_MAPPINGS = {
|
|
"RealtimeLoraTrainer": "Realtime LoRA Trainer",
|
|
"ApplyTrainedLora": "Apply Trained LoRA",
|
|
"SDXLLoraTrainer": "Realtime LoRA Trainer (SDXL - sd-scripts)",
|
|
"SD15LoraTrainer": "Realtime LoRA Trainer (SD 1.5 - sd-scripts)",
|
|
"MusubiZImageLoraTrainer": "Realtime LoRA Trainer (Z-Image - Musubi Tuner)",
|
|
"MusubiQwenImageLoraTrainer": "Realtime LoRA Trainer (Qwen Image - Musubi Tuner)",
|
|
"MusubiQwenImageEditLoraTrainer": "Realtime LoRA Trainer (Qwen Image Edit - Musubi Tuner)",
|
|
"MusubiWanLoraTrainer": "Realtime LoRA Trainer (Wan 2.2 - Musubi Tuner)",
|
|
"LoRALoaderWithAnalysis": "LoRA Loader + Analyzer",
|
|
"SDXLSelectiveLoRALoader": "Selective LoRA Loader (SDXL)",
|
|
"ZImageSelectiveLoRALoader": "Selective LoRA Loader (Z-Image)",
|
|
"FLUXSelectiveLoRALoader": "Selective LoRA Loader (FLUX)",
|
|
"WanSelectiveLoRALoader": "Selective LoRA Loader (Wan)",
|
|
"QwenSelectiveLoRALoader": "Selective LoRA Loader (Qwen)",
|
|
}
|
|
|
|
__all__ = ['NODE_CLASS_MAPPINGS', 'NODE_DISPLAY_NAME_MAPPINGS', 'WEB_DIRECTORY']
|