Delete assets/nodes/ConditionalLoRAApplier.py
This commit is contained in:
Creepybits
@@ -1,159 +0,0 @@
|
||||
import os
|
||||
import folder_paths
|
||||
from nodes import LoraLoader # This is correctly imported now
|
||||
|
||||
|
||||
def get_recursive_filenames(folder_name):
|
||||
full_path_dir = folder_paths.get_folder_paths(folder_name)
|
||||
if not full_path_dir:
|
||||
return []
|
||||
|
||||
filenames = []
|
||||
for base_dir in full_path_dir:
|
||||
for root, dirs, files in os.walk(base_dir):
|
||||
for file in files:
|
||||
# Construct full path to file
|
||||
full_file_path = os.path.join(root, file)
|
||||
# Get relative path from base_dir and normalize slashes
|
||||
relative_path = os.path.relpath(full_file_path, base_dir).replace("\\", "/")
|
||||
filenames.append(relative_path)
|
||||
return filenames
|
||||
|
||||
class ConditionalLoRAApplier:
|
||||
def __init__(self):
|
||||
pass
|
||||
|
||||
@classmethod
|
||||
def INPUT_TYPES(s):
|
||||
lora_list = get_recursive_filenames("loras")
|
||||
lora_names = ["None"] + [l for l in lora_list if l.lower().endswith((".safetensors", ".ckpt"))]
|
||||
lora_names.sort(key=lambda x: x.lower()) # Ensure sorted list for consistent dropdown
|
||||
|
||||
return {
|
||||
"required": {
|
||||
"model": ("MODEL",),
|
||||
"clip": ("CLIP",),
|
||||
"prompt": ("STRING", {"multiline": True, "default": ""}),
|
||||
"lora_definitions": ("STRING", {
|
||||
"multiline": True,
|
||||
"default": """
|
||||
# Define your LoRA rules here.
|
||||
# Format: keyword_phrase_1, keyword_phrase_2, ... : lora_full_relative_path, lora_strength, clip_strength
|
||||
# If ANY of the comma-separated keyword phrases are found in the prompt, the LoRA will be applied.
|
||||
# Example (use forward slashes for paths):
|
||||
# portrait, face detail: Flux/Details/amateur_photo_v1.safetensors, 0.75, 1.0
|
||||
# cinematic scene, movie shot: MyLoRAs/Styles/retro_cinematic_v2.safetensors, 0.8, 0.9
|
||||
# fantasy creature, mythical beast: Custom/Creatures/mythic_beast_lora.safetensors, 0.9, 0.9
|
||||
# Use comma-separated values for strength. Default is 1.0 if omitted.
|
||||
# Keep strength between -2.0 and 2.0.
|
||||
# All keyword_phrases should be found anywhere in the prompt (case-insensitive by default).
|
||||
"""
|
||||
}),
|
||||
},
|
||||
"optional": {
|
||||
"default_lora_name": (lora_names, {"default": "None"}),
|
||||
"default_lora_strength": ("FLOAT", {"default": 1.0, "min": -2.0, "max": 2.0, "step": 0.01}),
|
||||
"default_clip_strength": ("FLOAT", {"default": 1.0, "min": -2.0, "max": 2.0, "step": 0.01}),
|
||||
"case_sensitive": ("BOOLEAN", {"default": False}),
|
||||
}
|
||||
}
|
||||
|
||||
RETURN_TYPES = ("MODEL", "CLIP",)
|
||||
RETURN_NAMES = ("MODEL", "CLIP",)
|
||||
|
||||
FUNCTION = "apply_conditional_lora"
|
||||
CATEGORY = "Creepybits/Model Patcher"
|
||||
|
||||
def apply_conditional_lora(self, model, clip, prompt, lora_definitions, default_lora_name, default_lora_strength, default_clip_strength, case_sensitive):
|
||||
loras_to_apply = []
|
||||
|
||||
processed_prompt = prompt if case_sensitive else prompt.lower()
|
||||
|
||||
rules = lora_definitions.strip().split('\n')
|
||||
for rule_line in rules:
|
||||
rule_line = rule_line.strip()
|
||||
if not rule_line or rule_line.startswith('#'):
|
||||
continue
|
||||
|
||||
try:
|
||||
parts = rule_line.split(':', 1)
|
||||
if len(parts) < 2:
|
||||
print(f"Warning: Malformed LoRA rule (missing colon): '{rule_line}'")
|
||||
continue
|
||||
|
||||
keyword_string = parts[0].strip()
|
||||
keywords_for_this_lora = [k.strip() for k in keyword_string.split(',')]
|
||||
if not case_sensitive:
|
||||
keywords_for_this_lora = [k.lower() for k in keywords_for_this_lora]
|
||||
|
||||
match_found = False
|
||||
for kw in keywords_for_this_lora:
|
||||
if kw and kw in processed_prompt: # Added 'kw and' to prevent matching empty string
|
||||
match_found = True
|
||||
break
|
||||
|
||||
if match_found:
|
||||
# FIX: Changed 'lora_info_str' to 'parts[1]'
|
||||
lora_details = [x.strip() for x in parts[1].split(',', 2)]
|
||||
|
||||
filename = lora_details[0]
|
||||
strength = float(lora_details[1]) if len(lora_details) > 1 else 1.0
|
||||
clip_strength = float(lora_details[2]) if len(lora_details) > 2 else 1.0
|
||||
|
||||
# Use os.path.join for robustness, though folder_paths.get_full_path should handle it
|
||||
# Also added more specific warning if the file isn't found
|
||||
full_lora_path = folder_paths.get_full_path("loras", filename)
|
||||
if full_lora_path is None:
|
||||
print(f"Warning: LoRA file '{filename}' not found in 'loras' directory (or subdirectories) for rule '{rule_line}'. Skipping this rule.")
|
||||
continue
|
||||
|
||||
loras_to_apply.append((filename, strength, clip_strength))
|
||||
|
||||
except ValueError as e:
|
||||
print(f"Warning: Error parsing numeric strength in rule '{rule_line}': {e}. Skipping.")
|
||||
except IndexError as e: # Catch cases where split might not give enough parts if commas are missing
|
||||
print(f"Warning: Malformed strength/clip_strength format in rule '{rule_line}': {e}. Skipping.")
|
||||
except Exception as e:
|
||||
print(f"An unexpected error occurred while parsing rule '{rule_line}': {type(e).__name__} - {e}. Skipping.")
|
||||
|
||||
final_model = model
|
||||
final_clip = clip
|
||||
|
||||
# Apply default LoRA only if no conditional LoRAs were found
|
||||
if not loras_to_apply:
|
||||
if default_lora_name and default_lora_name != "None":
|
||||
full_default_lora_path = folder_paths.get_full_path("loras", default_lora_name)
|
||||
if full_default_lora_path is None:
|
||||
print(f"Warning: Default LoRA file '{default_lora_name}' not found. Skipping default LoRA application.")
|
||||
else:
|
||||
loras_to_apply.append((default_lora_name, default_lora_strength, default_clip_strength))
|
||||
|
||||
if not loras_to_apply:
|
||||
print("No LoRA selected or found for the given prompt/rules/default. Returning original model and clip.")
|
||||
return (final_model, final_clip,)
|
||||
|
||||
lora_loader_instance = LoraLoader()
|
||||
|
||||
for lora_filename, lora_strength, clip_strength in loras_to_apply:
|
||||
try:
|
||||
# The LoraLoader handles finding the full path internally, just needs the name
|
||||
final_model, final_clip = lora_loader_instance.load_lora(
|
||||
model=final_model,
|
||||
clip=final_clip,
|
||||
lora_name=lora_filename, # This expects just the filename/relative path, not full path
|
||||
strength_model=lora_strength,
|
||||
strength_clip=clip_strength
|
||||
)
|
||||
print(f"Applied LoRA: '{lora_filename}' (Model: {lora_strength}, CLIP: {clip_strength})") # Log success
|
||||
except Exception as e:
|
||||
print(f"FATAL ERROR applying LoRA '{lora_filename}': {type(e).__name__} - {e}. Skipping this LoRA.")
|
||||
|
||||
return (final_model, final_clip,)
|
||||
|
||||
NODE_CLASS_MAPPINGS = {
|
||||
"ConditionalLoRAApplier": ConditionalLoRAApplier,
|
||||
}
|
||||
|
||||
NODE_DISPLAY_NAME_MAPPINGS = {
|
||||
"ConditionalLoRAApplier": "Conditional LoRA Applier (Creepybits)",
|
||||
}
|
||||
Reference in New Issue
Block a user