"""
Ultimate Config Builder - Complete HTML UI Version
ALL data stored in single widget (lora_config)
Python reads everything from that widget
"""
import os
import sys
import json
import time
import builtins
import folder_paths
from typing import List, Dict, Any
import server
from aiohttp import web
import hashlib
from .civitai import civitai_fetch_by_hash
def safe_print(*args, **kwargs):
"""
Windows-safe print that survives colorama/stdout corruption.
Falls back to raw sys.__stdout__ if the wrapped stdout is broken,
and silently drops the message if even that fails. This prevents
OSError [Errno 22] from Windows console bugs crashing node execution.
"""
try:
builtins.print(*args, **kwargs)
except (OSError, ValueError):
try:
msg = " ".join(str(a) for a in args) + kwargs.get("end", "\n")
sys.__stdout__.write(msg)
sys.__stdout__.flush()
except Exception:
pass # Drop silently — don't crash the node over a log line
# Shadow the module-level `print` so every print() call in this file routes
# through safe_print. Without this, sibling methods (process_lora_array,
# lookup_triggers_endpoint, etc.) still hit the broken stdout and crash with
# OSError [Errno 22] on Windows — the local `print = safe_print` in
# generate_config only covers that one function.
print = safe_print
# =============================================================================
# INLINE LORA_UTILS FUNCTIONS (for compatibility)
# =============================================================================
def load_json_from_file(file_path):
"""Load JSON data from file"""
try:
with open(file_path, 'r') as json_file:
data = json.load(json_file)
return data
except FileNotFoundError:
return None
except json.JSONDecodeError:
print(f"[ConfigBuilder] Error decoding JSON in file: {file_path}")
return None
def save_dict_to_json(data_dict, file_path):
"""Save dictionary to JSON file"""
try:
# Ensure directory exists
os.makedirs(os.path.dirname(file_path), exist_ok=True)
with open(file_path, 'w') as json_file:
json.dump(data_dict, json_file, indent=4)
print(f"[ConfigBuilder] Data saved to {file_path}")
except Exception as e:
print(f"[ConfigBuilder] Error saving JSON to file: {e}")
def calculate_sha256(file_path):
"""Calculate SHA256 hash of a file"""
sha256_hash = hashlib.sha256()
with open(file_path, "rb") as f:
for chunk in iter(lambda: f.read(4096), b""):
sha256_hash.update(chunk)
return sha256_hash.hexdigest()
def get_model_version_info(hash_value):
"""Fetch model version info from Civitai API using hash"""
return civitai_fetch_by_hash(hash_value)
def load_and_save_tags(lora_name, force_fetch=False, auto_fetch=True):
"""
Load trigger tags for a LoRA, fetching from Civitai API if necessary.
Caches results to loras_tags.json.
Args:
lora_name: Name of the LoRA file (may include path with / or \\)
force_fetch: Force fetch from API even if cached
auto_fetch: If False, don't hash uncached LoRAs (just save empty entry)
Returns:
List of trigger words/tags
"""
output_dir = folder_paths.get_output_directory()
json_tags_path = os.path.join(output_dir, "benchmarks", "loras_tags.json")
lora_tags = load_json_from_file(json_tags_path)
# Normalize the lora_name to use forward slashes for cache lookup
normalized_name = lora_name.replace("\\", "/")
# Try to find in cache
output_tags = None
found_in_cache = False
if lora_tags is not None:
# Try original name first
if lora_name in lora_tags:
output_tags = lora_tags[lora_name]
found_in_cache = True
# If not found, try normalized name
elif normalized_name != lora_name and normalized_name in lora_tags:
output_tags = lora_tags[normalized_name]
found_in_cache = True
# If still not found, try backslash version
else:
backslash_name = normalized_name.replace("/", "\\")
if backslash_name in lora_tags:
output_tags = lora_tags[backslash_name]
found_in_cache = True
# If found in cache, return immediately
if found_in_cache:
return output_tags if output_tags is not None else []
# NOT IN CACHE - Need to fetch from API
lora_path = folder_paths.get_full_path("loras", lora_name)
# Check if lora_path is valid before attempting to hash
if lora_path is None:
print(f"[ConfigBuilder] ⚠️ LoRA file not found: {lora_name}")
return []
# Only hash if force_fetch is True OR (not in cache AND auto_fetch is True)
if force_fetch or auto_fetch:
print(f"[ConfigBuilder] 🔄 Fetching tags for {lora_name}")
try:
LORAsha256 = calculate_sha256(lora_path)
model_info = get_model_version_info(LORAsha256)
if model_info is not None and "trainedWords" in model_info:
print(f"[ConfigBuilder] ✅ Tags found for {lora_name}")
if lora_tags is None:
lora_tags = {}
lora_tags[normalized_name] = model_info["trainedWords"]
save_dict_to_json(lora_tags, json_tags_path)
return model_info["trainedWords"]
else:
print(f"[ConfigBuilder] No tags found for {lora_name}")
if lora_tags is None:
lora_tags = {}
lora_tags[normalized_name] = []
save_dict_to_json(lora_tags, json_tags_path)
return []
except Exception as e:
print(f"[ConfigBuilder] Error processing {lora_name}: {e}")
return []
else:
# Not in cache and auto_fetch=False
if lora_tags is None:
lora_tags = {}
lora_tags[normalized_name] = []
save_dict_to_json(lora_tags, json_tags_path)
return []
class UltimateConfigBuilder:
"""
Config builder with complete HTML UI.
All data is stored in the lora_config widget as a single JSON object.
"""
@classmethod
def INPUT_TYPES(cls):
sessions = cls.get_available_sessions()
return {
"required": {
# Session Management (hidden, controlled by HTML)
"session_name": ("STRING", {
"default": "my_test_session",
"multiline": False
}),
"load_session": (sessions, {
"default": sessions[0] if sessions else "None"
}),
# Sampler Settings (hidden, controlled by HTML)
"samplers": ("STRING", {
"default": "euler, dpmpp_2m",
"multiline": False
}),
"schedulers": ("STRING", {
"default": "normal, karras",
"multiline": False
}),
"steps": ("STRING", {
"default": "20, 30",
"multiline": False
}),
"cfg": ("STRING", {
"default": "7.0",
"multiline": False
}),
# LoRA Configuration (ACTUAL DATA STORAGE - contains EVERYTHING)
"lora_config": ("STRING", {
"default": cls.get_default_config(),
"multiline": True
}),
# Options (hidden, controlled by HTML)
"include_none": ("BOOLEAN", {
"default": False
}),
},
"optional": {
"model": ("STRING", {
"default": "",
"multiline": False
}),
}
}
# session_name is carried through configs_json's _session_settings now,
# so the dedicated output socket is redundant. Removed to simplify the
# node interface.
RETURN_TYPES = ("STRING",)
RETURN_NAMES = ("configs_json",)
FUNCTION = "generate_config"
CATEGORY = "sampling/testing"
OUTPUT_NODE = True
@staticmethod
def get_default_config():
"""Return default complete configuration"""
config = {
"session_name": "my_test_session",
"include_none": False,
"global_positive_groups": [],
"global_negative": "",
"config_arrays": [
{
"name": "Config 1",
"samplers": "euler, dpmpp_2m",
"schedulers": "normal, karras",
"steps": "20, 30",
"cfg": "7.0",
"model": "",
"loras": ["None"],
"lora_omit_triggers": [],
"lora_triggerwords_append_settings": {},
"combine": True,
"positive_prompt_groups": [],
"negative_prompt": "",
"use_custom_prompts": False
}
]
}
return json.dumps(config, indent=2, ensure_ascii=False)
@staticmethod
def get_available_sessions() -> List[str]:
"""Scan benchmarks folder for available sessions, sorted by newest first"""
sessions = ["None"]
try:
output_dir = folder_paths.get_output_directory()
benchmarks_dir = os.path.join(output_dir, "benchmarks")
if os.path.exists(benchmarks_dir):
session_items = []
for item in os.listdir(benchmarks_dir):
item_path = os.path.join(benchmarks_dir, item)
manifest_path = os.path.join(item_path, "manifest.json")
if os.path.isdir(item_path) and os.path.exists(manifest_path):
# Get modification time for sorting
mtime = os.path.getmtime(item_path)
session_items.append((item, mtime))
# Sort by modification time (newest first)
session_items.sort(key=lambda x: x[1], reverse=True)
sessions.extend([item[0] for item in session_items])
except Exception as e:
print(f"[ConfigBuilder] Warning: Could not scan sessions: {e}")
return sessions
@staticmethod
def expand_lora_folders(lora_list: List[str]) -> List[str]:
"""
Expand folder references to individual LoRA files.
Args:
lora_list: List of LoRA strings (may include folders)
Returns:
List with folders expanded to individual files
"""
expanded = []
available_loras = folder_paths.get_filename_list("loras")
for lora_str in lora_list:
if not lora_str or lora_str == "None":
continue
# Parse out the name (before any : strength modifiers)
lora_name = lora_str.split(":")[0]
# Check if it's a folder reference
is_folder = lora_name.endswith("/") or lora_name.endswith("/*")
if is_folder:
# Remove trailing / or /*
folder_name = lora_name.rstrip("/*").rstrip("/")
# Normalize to forward slashes for comparison
folder_prefix = folder_name.replace("\\", "/") + "/"
# Find all LoRAs in this folder
for lora_file in available_loras:
normalized_file = lora_file.replace("\\", "/")
if normalized_file.startswith(folder_prefix):
expanded.append(lora_file)
lora_count = len([l for l in available_loras if l.replace('\\', '/').startswith(folder_prefix)])
print(f"[ConfigBuilder] Expanded folder '{lora_name}' to {lora_count} LoRAs")
else:
# Regular LoRA file
expanded.append(lora_str)
return expanded
@staticmethod
def lookup_lora_triggers(lora_list: List[str]) -> Dict[str, List[str]]:
"""
Lookup trigger words for a list of LoRAs.
Args:
lora_list: List of LoRA strings (may include strengths and folders)
Returns:
Dict mapping lora name to list of trigger words
"""
# First, expand any folder references
expanded_list = UltimateConfigBuilder.expand_lora_folders(lora_list)
trigger_map = {}
for lora_str in expanded_list:
if not lora_str or lora_str == "None":
continue
# Handle combined LoRAs (e.g., "lora1 + lora2")
if " + " in lora_str:
parts = lora_str.split(" + ")
for part in parts:
part = part.strip()
if part and part != "None":
lora_name = part.split(":")[0]
if lora_name not in trigger_map:
try:
triggers = load_and_save_tags(
lora_name,
force_fetch=False,
auto_fetch=True
)
trigger_map[lora_name] = triggers if triggers else []
except Exception as e:
print(f"[ConfigBuilder] Error fetching triggers for {lora_name}: {e}")
trigger_map[lora_name] = []
else:
# Single LoRA
lora_name = lora_str.split(":")[0]
if lora_name and lora_name != "None" and lora_name not in trigger_map:
try:
triggers = load_and_save_tags(
lora_name,
force_fetch=False,
auto_fetch=True
)
trigger_map[lora_name] = triggers if triggers else []
except Exception as e:
print(f"[ConfigBuilder] Error fetching triggers for {lora_name}: {e}")
trigger_map[lora_name] = []
return trigger_map
@staticmethod
def parse_int_list(value: str) -> List[int]:
"""Parse comma-separated integers"""
items = UltimateConfigBuilder.parse_comma_list(value)
result = []
for item in items:
try:
# Cast to float first to handle strings like "20.0", then to int
result.append(int(float(item)))
except ValueError:
print(f"[ConfigBuilder] Warning: Could not parse integer '{item}'")
return result
@staticmethod
def parse_comma_list(value) -> List[str]:
"""Parse comma-separated string or pass through list"""
if isinstance(value, list):
return [str(item).strip() for item in value if str(item).strip()]
if not value or str(value).strip() == "":
return []
return [item.strip() for item in str(value).split(",") if item.strip()]
@staticmethod
def parse_number_list(value: str) -> List[float]:
"""Parse comma-separated numbers"""
items = UltimateConfigBuilder.parse_comma_list(value)
result = []
for item in items:
try:
result.append(float(item))
except ValueError:
print(f"[ConfigBuilder] Warning: Could not parse '{item}'")
return result
@staticmethod
def process_lora_array(config_array: Dict, include_none: bool) -> List[str]:
"""
Process a SINGLE config array and return its lora strings.
Args:
config_array: Single config array dict from config_arrays
include_none: Whether to include "None" in results
Returns:
List of lora strings for this config array
"""
array_name = config_array.get("name", "Unnamed Config")
# FIX: Force combine to True.
# The UI defaults new configs to 'combine: false' but the Preview
# treats them as combined. We enforce True here to match the Preview/Stack behavior.
combine = True
loras = config_array.get("loras", [])
lora_bypass_states = config_array.get("lora_bypass_states", {})
# Filter out bypassed loras, then convert to strings
lora_strings = []
for lora in loras:
if not lora or lora == "None":
continue
lora_str = str(lora)
# Extract lora name (path before first colon) to check bypass state
lora_name = lora_str.split(":")[0] if ":" in lora_str else lora_str
if lora_bypass_states.get(lora_name, False):
continue # Skip bypassed
lora_strings.append(lora_str)
# Apply weight arrays (bracket notation) for "Compare Strengths" entries.
# The Builder UI stores per-LoRA strength arrays in a side-channel
# (config_array.lora_weight_arrays[name + "_model" / "_clip"]). Rewrite
# each lora string from "name:m:c" -> "name:[m_arr]:[c_arr|c]" so the
# orchestrator's _expand_lora_weight_arrays can fan them out at runtime.
# SYNC WARNING: must match convertStateToConfigs() in conf-builder-utilities.js.
weight_arrays = config_array.get("lora_weight_arrays", {}) or {}
def _apply_weight_array(lora_str: str) -> str:
name, _sep, rest = lora_str.partition(":")
parts = rest.split(":") if rest else []
model_str = parts[0] if parts else "1.00"
clip_str = parts[1] if len(parts) > 1 else model_str
model_arr = weight_arrays.get(name + "_model")
clip_arr = weight_arrays.get(name + "_clip")
if model_arr and len(model_arr) > 1:
model_part = "[" + ", ".join(str(v) for v in model_arr) + "]"
if clip_arr and len(clip_arr) > 1:
# Legacy/edge case: separate clip array → preserves Cartesian behavior
# for backward-compat with workflows saved before the locked-semantics
# simplification (2026-04-28).
clip_part = "[" + ", ".join(str(v) for v in clip_arr) + "]"
return f"{name}:{model_part}:{clip_part}"
# Locked semantics: omit the clip part entirely.
# parse_lora_definition (config_utils.py) defaults clip = model when no
# third segment is present, so "name:[a,b,c]" expands to N configs each
# with model=clip=value. This avoids the Cartesian explosion that
# "name:[m]:[c]" would trigger in expand_lora_stack.
return f"{name}:{model_part}"
if clip_arr and len(clip_arr) > 1:
# Edge: clip-only array (no model array) — emit as before.
return f"{name}:{model_str}:[" + ", ".join(str(v) for v in clip_arr) + "]"
return lora_str
lora_strings = [_apply_weight_array(s) for s in lora_strings]
# Add combined version if requested
if combine and len(lora_strings) > 1:
# Folder refs ("name/") are excluded from stacking — they expand at
# runtime into multiple loras. Detect by the NAME portion (before the
# first colon) ending in "/", not the whole string, so that
# "folder/:[a,b]:[c,d]" (folder ref + strength array) is also excluded.
stackable = [s for s in lora_strings if not s.split(":", 1)[0].endswith("/")]
if len(stackable) > 1:
# When combine is true, ONLY return the combined version
combined = " + ".join(stackable)
lora_strings = [combined]
print(f"[ConfigBuilder] {array_name}: Combined {len(stackable)} LoRAs into stack")
# Remove duplicates while preserving order
seen = set()
unique_strings = []
for item in lora_strings:
if item not in seen:
seen.add(item)
unique_strings.append(item)
print(f"[ConfigBuilder] {array_name}: Processed {len(unique_strings)} LoRA configs")
return unique_strings
@staticmethod
def state_to_configs_json(state: dict) -> str:
"""
Pure transformer: builder UI state -> configs_json string.
Single source of truth. Called by:
- generate_config() (run-time, via lora_config widget)
- /configbuilder/preview (preview endpoint, via POST body)
Both code paths share this function. By construction, the preview
and the actual node output cannot disagree.
Args:
state: parsed builder UI state (the JSON object stored in the
lora_config widget — config_arrays, prompts, etc.)
Returns:
The configs_json string (top-level: {"configs": [...], "_distribution": ..., "_session_settings": ...}).
"""
# SINGLE SOURCE OF TRUTH for builder UI state -> configs_json.
actual_include_none = state.get("include_none", False)
config_arrays = state.get("config_arrays", [])
# Global prompts (used when per-config prompts are not defined)
global_positive_groups = state.get("global_positive_groups", [])
global_negative = state.get("global_negative", "")
if not config_arrays:
config_arrays = [{
"name": "Config 1",
"samplers": ["euler"],
"schedulers": ["normal"],
"steps": "20",
"cfg": "7.0",
"model": "",
"loras": ["None"],
"lora_omit_triggers": [],
"lora_triggerwords_append_settings": {},
"combine": True,
"positive_prompt_groups": [],
"negative_prompt": "",
"use_custom_prompts": False
}]
configs_output = []
total_lora_configs = 0
for config_array in config_arrays:
# Parse values from this config array
sampler_list = UltimateConfigBuilder.parse_comma_list(config_array.get("samplers", "euler"))
scheduler_list = UltimateConfigBuilder.parse_comma_list(config_array.get("schedulers", "normal"))
steps_list = UltimateConfigBuilder.parse_int_list(config_array.get("steps", "20"))
cfg_list = UltimateConfigBuilder.parse_number_list(config_array.get("cfg", "7.0"))
models_raw = config_array.get("models", ["None"])
omit_triggers = config_array.get("lora_omit_triggers", [])
lora_triggerwords_append_settings = config_array.get("lora_triggerwords_append_settings", {})
# Extra Model & Sampling Options
model_sampling_override = config_array.get("model_sampling_override", "none")
model_sampling_shift = config_array.get("model_sampling_shift", "1.73")
model_sampling_flux_max_shift = config_array.get("model_sampling_flux_max_shift", "1.15")
model_sampling_flux_base_shift = config_array.get("model_sampling_flux_base_shift", "0.5")
use_advanced_sampling = config_array.get("use_advanced_sampling", False)
advanced_guider = config_array.get("advanced_guider", "cfg_guider")
advanced_scheduler = config_array.get("advanced_scheduler", "basic")
use_flux_guidance = config_array.get("use_flux_guidance", False)
flux_guidance_value = config_array.get("flux_guidance_value", "3.5")
# Process models - handle both object format {path, type} and legacy string format
model_strings = []
model_type = "checkpoint" # default
model_bypass_states = config_array.get("model_bypass_states", {})
for m in models_raw:
if isinstance(m, dict):
# Trust the user's type selection even if no file is picked yet
# (otherwise switching to LTX/diffusion_model type without yet
# selecting a file silently falls back to checkpoint).
if m.get("type"):
model_type = m.get("type", "checkpoint")
path = m.get("path", "")
if path and path != "None" and not model_bypass_states.get(path, False):
model_strings.append(str(path))
elif isinstance(m, str) and m and m != "None" and not model_bypass_states.get(m, False):
model_strings.append(str(m))
# Process loras for this config
lora_strings = UltimateConfigBuilder.process_lora_array(config_array, actual_include_none)
total_lora_configs += len(lora_strings)
# Create ONE config for this array
config = {
"sampler": sampler_list if len(sampler_list) > 1 else sampler_list[0] if sampler_list else "euler",
"scheduler": scheduler_list if len(scheduler_list) > 1 else scheduler_list[0] if scheduler_list else "normal",
"steps": steps_list if len(steps_list) > 1 else steps_list[0] if steps_list else 20,
"cfg": cfg_list if len(cfg_list) > 1 else cfg_list[0] if cfg_list else 7.0,
"lora": lora_strings if len(lora_strings) > 1 else lora_strings[0] if lora_strings else "None",
"model": model_strings if len(model_strings) > 1 else model_strings[0] if model_strings else "None"
}
# Always include all fields with defaults to prevent manifest data loss
# when fields are toggled off in the UI and the session is reloaded
seed_behavior = config_array.get("seed_behavior", "fixed")
config["seed_behavior"] = seed_behavior
# Full run seed behavior (applied before/after entire grid test session)
full_run_seed_behavior = config_array.get("full_run_seed_behavior", "fixed")
config["full_run_seed_behavior"] = full_run_seed_behavior
# Full run seed (overrides node seed when > 0)
full_run_seed = config_array.get("full_run_seed", 0)
config["full_run_seed"] = int(full_run_seed) if full_run_seed else 0
# Process VAEs — filter out bypassed (unchecked) entries AND any
# empty / "None" placeholders. If nothing remains, OMIT the "vae"
# key entirely so the orchestrator falls back to the default
# behavior (no per-config VAE override). Emitting "vae": "None"
# made the generator try to load a model file literally named
# "None", which fails.
vaes_raw = config_array.get("vaes", ["None"])
vae_bypass_states = config_array.get("vae_bypass_states", {}) or {}
vae_strings = [
str(v) for v in vaes_raw
if v and v != "None" and not vae_bypass_states.get(str(v), False)
]
if vae_strings:
config["vae"] = vae_strings if len(vae_strings) > 1 else vae_strings[0]
# Always include model_type and related fields
config["model_type"] = model_type
if model_type != "checkpoint":
text_encoders = config_array.get("text_encoders", [])
if text_encoders:
config["text_encoders"] = [te for te in text_encoders if te and te != "None"]
clip_type = config_array.get("clip_type", "")
if clip_type:
config["clip_type"] = clip_type
if model_type == "gguf":
gguf_options = config_array.get("gguf_options", {})
if gguf_options:
config["gguf_options"] = gguf_options
if model_type == "ltx_video":
# Flatten configArray.ltx_video into top-level fields for the orchestrator.
# Mirrors the JS-side flattening in conf-builder-utilities.convertStateToConfigs.
ltx = config_array.get("ltx_video") or {}
if ltx.get("clip_models"):
config["clip_models"] = ltx["clip_models"]
if ltx.get("vae_video"):
config["vae_video"] = ltx["vae_video"]
if ltx.get("vae_audio"):
config["vae_audio"] = ltx["vae_audio"]
if ltx.get("latent_upscaler"):
config["latent_upscaler"] = ltx["latent_upscaler"]
if ltx.get("duration_seconds") is not None:
config["duration_seconds"] = ltx["duration_seconds"]
if ltx.get("frame_rate") is not None:
config["frame_rate"] = ltx["frame_rate"]
if ltx.get("sampler_stage1"):
config["sampler_stage1"] = ltx["sampler_stage1"]
if ltx.get("sampler_stage2"):
config["sampler_stage2"] = ltx["sampler_stage2"]
if ltx.get("sigmas_stage1"):
config["sigmas_stage1"] = ltx["sigmas_stage1"]
if ltx.get("sigmas_stage2"):
config["sigmas_stage2"] = ltx["sigmas_stage2"]
if ltx.get("input_image") is not None:
config["input_image"] = ltx["input_image"]
if ltx.get("image_strength_stage1") is not None:
config["image_strength_stage1"] = ltx["image_strength_stage1"]
if ltx.get("image_strength_stage2") is not None:
config["image_strength_stage2"] = ltx["image_strength_stage2"]
if ltx.get("img_compression") is not None:
config["img_compression"] = ltx["img_compression"]
if ltx.get("audio_mode"):
config["audio_mode"] = ltx["audio_mode"]
# Always include omit triggers (empty list if none)
config["lora_omit_triggers"] = omit_triggers if omit_triggers else []
# Always include trigger append settings (empty dict if none)
config["lora_triggerwords_append_settings"] = lora_triggerwords_append_settings if lora_triggerwords_append_settings else {}
# Per-config resolutions (override sampler's resolutions_json)
raw_resolutions = config_array.get("resolutions", [])
if raw_resolutions and len(raw_resolutions) > 0:
# Convert "WxH" strings to [W, H] arrays for config_utils.expand_configs()
parsed_res = []
for r in raw_resolutions:
if isinstance(r, str) and "x" in r:
parts = r.split("x")
parsed_res.append([int(parts[0]), int(parts[1])])
elif isinstance(r, (list, tuple)) and len(r) == 2:
parsed_res.append([int(r[0]), int(r[1])])
if parsed_res:
config["resolutions"] = parsed_res
# Attention mode(s) for testing different attention implementations
attention_modes = config_array.get("attention_modes", ["default"])
if isinstance(attention_modes, list):
filtered = [a for a in attention_modes if a and a != "default"]
if filtered:
config["attention_mode"] = filtered if len(filtered) > 1 else filtered[0]
else:
config["attention_mode"] = "default"
elif isinstance(attention_modes, str):
config["attention_mode"] = attention_modes
else:
config["attention_mode"] = "default"
# Model prompt prefix/suffix (quality tags prepended/appended to prompts)
model_prompt_prefix = config_array.get("model_prompt_prefix", "")
config["model_prompt_prefix"] = model_prompt_prefix.strip() if model_prompt_prefix else ""
model_prompt_suffix = config_array.get("model_prompt_suffix", "")
config["model_prompt_suffix"] = model_prompt_suffix.strip() if model_prompt_suffix else ""
# Always include model sampling options with defaults
config["model_sampling_override"] = model_sampling_override if model_sampling_override else "none"
if model_sampling_override and model_sampling_override != "none":
if model_sampling_override == "flux":
config["model_sampling_flux_max_shift"] = model_sampling_flux_max_shift
config["model_sampling_flux_base_shift"] = model_sampling_flux_base_shift
elif model_sampling_override == "flux2":
config["model_sampling_shift"] = model_sampling_shift if model_sampling_shift else "2.02"
else:
config["model_sampling_shift"] = model_sampling_shift
config["use_advanced_sampling"] = use_advanced_sampling or False
if use_advanced_sampling:
config["advanced_guider"] = advanced_guider
config["advanced_scheduler"] = advanced_scheduler
config["use_flux_guidance"] = use_flux_guidance or False
if use_flux_guidance:
config["flux_guidance_value"] = flux_guidance_value
# Deep Shrink (Kohya / PatchModelAddDownscale) — patches the UNet
# to downscale features at a specific block during early diffusion.
# Only emit detail params when toggle is on, to keep configs_json clean.
use_deep_shrink = config_array.get("use_deep_shrink", False)
config["use_deep_shrink"] = bool(use_deep_shrink)
if use_deep_shrink:
config["deep_shrink_block_number"] = int(config_array.get("deep_shrink_block_number", 3))
config["deep_shrink_downscale_factor"] = float(config_array.get("deep_shrink_downscale_factor", 2.0))
config["deep_shrink_start_percent"] = float(config_array.get("deep_shrink_start_percent", 0.0))
config["deep_shrink_end_percent"] = float(config_array.get("deep_shrink_end_percent", 0.35))
config["deep_shrink_downscale_after_skip"] = bool(config_array.get("deep_shrink_downscale_after_skip", True))
config["deep_shrink_downscale_method"] = str(config_array.get("deep_shrink_downscale_method", "bicubic"))
config["deep_shrink_upscale_method"] = str(config_array.get("deep_shrink_upscale_method", "bicubic"))
# ==== PROMPT HANDLING ====
# Priority: per-config > global > node inputs (omitted = use node inputs)
use_custom = config_array.get("use_custom_prompts", False)
per_config_positive_groups = config_array.get("positive_prompt_groups", [])
per_config_negative = config_array.get("negative_prompt", "")
if use_custom and per_config_positive_groups:
# Per-config prompts override everything
# Store as nested array format for parse_prompt_input_nested() compatibility
config["positive"] = per_config_positive_groups
if per_config_negative:
config["negative"] = per_config_negative
config["_prompt_source"] = "custom"
elif global_positive_groups:
# Global prompts override node inputs
config["positive"] = global_positive_groups
if global_negative:
config["negative"] = global_negative
config["_prompt_source"] = "global"
# If neither, omit "positive"/"negative" keys - node inputs will be used as fallback
# --- CONFIG SCHEMA VALIDATION ---
# Sanity check for typos / dropped fields. Only includes keys this
# function ALWAYS sets unconditionally — fields that are filled in
# later by the orchestrator (seed, denoise) or only set for
# specific model types (clip_type, text_encoders, gguf_options,
# ltx_*) are intentionally excluded so they don't fire false alarms.
_EXPECTED_CONFIG_KEYS = {
"sampler", "scheduler", "steps", "cfg", "seed_behavior",
"model", "model_type", "lora", "vae",
"model_sampling_override", "use_advanced_sampling", "use_flux_guidance",
"model_prompt_prefix", "model_prompt_suffix", "attention_mode",
"use_deep_shrink",
}
missing = _EXPECTED_CONFIG_KEYS - set(config.keys())
if missing:
print(f"[ConfigBuilder] ⚠️ Config schema drift: missing keys {missing} in config for '{config.get('model', '?')}'")
configs_output.append(config)
# Build the output object with configs and optional distribution settings
output_obj = {"configs": configs_output}
# Embed distribution config if enabled
if state.get("distribution_enabled") and state.get("worker_urls"):
output_obj["_distribution"] = {
"enabled": True,
"worker_urls": [u for u in state["worker_urls"] if u and u.strip()],
"claim_timeout": state.get("claim_timeout", 600),
"use_master_encoding": state.get("use_master_encoding", False),
"sync_models_to_workers": state.get("sync_models_to_workers", False)
}
# Embed session-level settings (upscaling, cooldown) if enabled
session_settings = {}
upscaling_data = state.get("upscaling", {})
if upscaling_data and upscaling_data.get("enabled", False):
# Filter out inactive pipelines and inactive steps within pipelines
pipelines = upscaling_data.get("pipelines", [])
active_pipelines = []
for p in pipelines:
if p.get("active", True) is False:
continue
active_steps = [s for s in p.get("steps", []) if s.get("active", True) is not False]
if active_steps:
active_pipelines.append({**p, "steps": active_steps})
if active_pipelines:
session_settings["upscaling"] = {
"enabled": True,
"save_pre_upscale": upscaling_data.get("save_pre_upscale", False),
"run_upscales_at_end": upscaling_data.get("run_upscales_at_end", False),
"hires_prompt_adjust": upscaling_data.get("hires_prompt_adjust", False),
"hires_prompt_behavior": upscaling_data.get("hires_prompt_behavior", "append_end"),
"hires_prompt_text": upscaling_data.get("hires_prompt_text", ""),
"pipelines": active_pipelines
}
cooldown_data = state.get("cooldown", {})
if cooldown_data and cooldown_data.get("enabled", False):
session_settings["cooldown"] = cooldown_data
# Start At Job # (skip to a specific job number) — always emit.
try:
session_settings["start_at_job"] = int(state.get("start_at_job", 0))
except (TypeError, ValueError):
session_settings["start_at_job"] = 0
# Image save format — always emit so Builder UI is authoritative.
session_settings["image_format"] = str(state.get("image_format", "webp"))
# Builder UI is authoritative for all run settings — emit every field.
# Type coercion is defensive; the Builder UI sends correct types but
# workflows saved before the Builder UI had these fields might have
# missing or string-typed values that need normalization.
session_settings["overwrite_existing"] = bool(state.get("overwrite_existing", False))
try:
session_settings["flush_batch_every"] = int(state.get("flush_batch_every", 1))
except (TypeError, ValueError):
session_settings["flush_batch_every"] = 1
session_settings["lora_triggerwords_mode"] = str(state.get("lora_triggerwords_mode", "None"))
session_settings["save_conditioning_cache_to_file"] = bool(state.get("save_conditioning_cache_to_file", False))
session_settings["enable_model_cache"] = bool(state.get("enable_model_cache", False))
try:
session_settings["vae_batch_size"] = int(state.get("vae_batch_size", 1))
except (TypeError, ValueError):
session_settings["vae_batch_size"] = 1
# session_name moves into session_settings too so the Generator's
# session_name widget can be removed (Phase 2).
session_settings["session_name"] = str(state.get("session_name", "my_session"))
try:
session_settings["add_random_seeds_to_gens"] = int(state.get("add_random_seeds_to_gens", 0))
except (TypeError, ValueError):
session_settings["add_random_seeds_to_gens"] = 0
if session_settings:
output_obj["_session_settings"] = session_settings
json_output = json.dumps(output_obj, indent=2, ensure_ascii=False)
return json_output
def generate_config(
self,
session_name,
load_session,
samplers,
schedulers,
steps,
cfg,
lora_config,
include_none,
model=""
):
"""
Generate configuration.
NOTE: All widget parameters are IGNORED!
The actual data comes from the lora_config widget which contains everything.
"""
# Use Windows-safe print for all logging in this function. Prevents
# OSError [Errno 22] from colorama/stdout corruption crashing the node.
print = safe_print
print(f"\n{'='*80}")
print(f"[ConfigBuilder] 🎯 Generating Configuration")
print(f"{'='*80}")
# Parse the COMPLETE state from lora_config widget
try:
state = json.loads(lora_config)
except json.JSONDecodeError as e:
print(f"[ConfigBuilder] ⚠️ Error parsing lora_config: {e}")
print(f"[ConfigBuilder] Using default config")
state = json.loads(self.get_default_config())
actual_session_name = state.get("session_name", session_name)
# ============================================================================
# SINGLE SOURCE OF TRUTH: state_to_configs_json is also called by
# POST /configbuilder/preview, so the preview cannot disagree with this.
# ============================================================================
json_output = UltimateConfigBuilder.state_to_configs_json(state)
# Brief summary log (computed from the helper's output)
try:
parsed = json.loads(json_output)
n_configs = len(parsed.get("configs", []))
except Exception:
n_configs = 0
print(f"[ConfigBuilder] 📊 Session: {actual_session_name}")
print(f"[ConfigBuilder] 📊 Configs: {n_configs}")
print(f"{'='*80}\n")
return (json_output,)
# API endpoint for trigger word lookup
@server.PromptServer.instance.routes.post("/configbuilder/lookup_triggers")
async def lookup_triggers_endpoint(request):
"""API endpoint to lookup trigger words for LoRAs"""
try:
from . import civitai as _civitai # local import keeps top of file clean
data = await request.json()
lora_list = data.get("loras", [])
print(f"[ConfigBuilder] 🔍 Lookup request for {len(lora_list)} LoRAs")
trigger_map = UltimateConfigBuilder.lookup_lora_triggers(lora_list)
print(f"[ConfigBuilder] ✅ Found triggers for {len(trigger_map)} LoRAs")
return web.json_response({
"triggers": trigger_map,
"civitai_available": _civitai.is_civitai_available(),
})
except Exception as e:
print(f"[ConfigBuilder] ❌ Error in lookup_triggers endpoint: {e}")
import traceback
traceback.print_exc()
return web.json_response({
"error": str(e)
}, status=500)
# API endpoint for detailed LoRA metadata lookup
@server.PromptServer.instance.routes.post("/configbuilder/lookup_lora_metadata")
async def lookup_lora_metadata_endpoint(request):
"""API endpoint to lookup full metadata for a specific LoRA from CivitAI"""
try:
data = await request.json()
lora_name = data.get("lora_name", "")
force_refresh = data.get("force_refresh", False)
if not lora_name:
return web.json_response({
"error": "No LoRA name provided"
}, status=400)
print(f"[ConfigBuilder] 🔍 Full metadata lookup request for: {lora_name}")
# Check disk cache first to avoid expensive SHA256 hashing
output_dir = folder_paths.get_output_directory()
model_data_dir = os.path.join(output_dir, "benchmarks", "model-data", lora_name.replace("/", "_").replace("\\", "_").replace(".safetensors", ""))
metadata_file = os.path.join(model_data_dir, "metadata.json")
if os.path.exists(metadata_file) and not force_refresh:
try:
cached = load_json_from_file(metadata_file)
if cached and cached.get("name"):
# Get file modification time for cache date display
cache_mtime = os.path.getmtime(metadata_file)
cache_date = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime(cache_mtime))
print(f"[ConfigBuilder] ✅ Using cached metadata for: {lora_name} (cached on {cache_date})")
return web.json_response({
"metadata": cached,
"saved_to": metadata_file,
"cached": True,
"cache_date": cache_date
})
except Exception:
pass # Cache miss or corrupt file - fall through to fresh lookup
# Get the full path to the LoRA file
lora_path = folder_paths.get_full_path("loras", lora_name)
if lora_path is None:
return web.json_response({
"error": f"LoRA file not found: {lora_name}"
}, status=404)
# Calculate the hash (expensive - only when not cached)
lora_hash = calculate_sha256(lora_path)
short_hash = lora_hash[:10] # First 10 characters for short hash
print(f"[ConfigBuilder] 📊 Hash calculated: {short_hash}")
# Fetch metadata from CivitAI
model_info = get_model_version_info(lora_hash)
if model_info is None:
return web.json_response({
"error": "No metadata found on CivitAI",
"hash": lora_hash,
"short_hash": short_hash
}, status=404)
# Extract relevant information
metadata = {
"name": model_info.get("name", "Unknown"),
"model_name": model_info.get("model", {}).get("name", "Unknown") if isinstance(model_info.get("model"), dict) else "Unknown",
"trained_words": model_info.get("trainedWords", []),
"base_model": model_info.get("baseModel", "Unknown"),
"description": model_info.get("description", ""),
"tags": model_info.get("model", {}).get("tags", []) if isinstance(model_info.get("model"), dict) else [],
"images": [],
"url": f"https://civitai.com/models/{model_info.get('modelId', '')}" if model_info.get("modelId") else "",
"hash": lora_hash,
"short_hash": short_hash,
"file_path": lora_path,
"stats": model_info.get("stats", {}),
"creator": model_info.get("creator", {}).get("username", "Unknown") if isinstance(model_info.get("creator"), dict) else "Unknown"
}
# Extract images
if "images" in model_info and isinstance(model_info["images"], list):
for img in model_info["images"][:5]: # Limit to first 5 images
if isinstance(img, dict) and "url" in img:
metadata["images"].append({
"url": img["url"],
"nsfw": img.get("nsfw", "None"),
"width": img.get("width", 0),
"height": img.get("height", 0)
})
# Save metadata to file
output_dir = folder_paths.get_output_directory()
model_data_dir = os.path.join(output_dir, "benchmarks", "model-data", lora_name.replace("/", "_").replace("\\", "_").replace(".safetensors", ""))
os.makedirs(model_data_dir, exist_ok=True)
metadata_file = os.path.join(model_data_dir, "metadata.json")
save_dict_to_json(metadata, metadata_file)
print(f"[ConfigBuilder] ✅ Metadata saved to: {metadata_file}")
# Compare fresh tags with stored loras_tags.json (only on force_refresh)
tags_changed = False
old_tags = []
new_tags = metadata.get("trained_words", [])
if force_refresh and new_tags:
json_tags_path = os.path.join(output_dir, "benchmarks/loras_tags.json")
if os.path.exists(json_tags_path):
lora_tags = load_json_from_file(json_tags_path) or {}
normalized = lora_name.replace("\\", "/")
backslash = lora_name.replace("/", "\\")
old_tags = lora_tags.get(lora_name, lora_tags.get(normalized, lora_tags.get(backslash, [])))
if old_tags is None:
old_tags = []
# Compare sorted lists to detect any difference
if sorted(old_tags) != sorted(new_tags):
tags_changed = True
print(f"[ConfigBuilder] ⚠️ Tags changed for {lora_name}: {old_tags} -> {new_tags}")
return web.json_response({
"metadata": metadata,
"saved_to": metadata_file,
"tags_changed": tags_changed,
"old_tags": old_tags,
"new_tags": new_tags
})
except Exception as e:
print(f"[ConfigBuilder] ❌ Error in lookup_lora_metadata endpoint: {e}")
import traceback
traceback.print_exc()
return web.json_response({
"error": str(e)
}, status=500)
# API endpoint for detailed Model/Checkpoint metadata lookup
@server.PromptServer.instance.routes.post("/configbuilder/lookup_model_metadata")
async def lookup_model_metadata_endpoint(request):
"""API endpoint to lookup full metadata for a model/checkpoint from CivitAI"""
try:
data = await request.json()
model_name = data.get("model_name", "")
model_type = data.get("model_type", "checkpoint")
force_refresh = data.get("force_refresh", False)
if not model_name:
return web.json_response({
"error": "No model name provided"
}, status=400)
print(f"[ConfigBuilder] 🔍 Full metadata lookup request for model: {model_name} (type: {model_type})")
# Check disk cache first to avoid expensive SHA256 hashing
output_dir = folder_paths.get_output_directory()
model_data_dir = os.path.join(output_dir, "benchmarks", "model-data", model_name.replace("/", "_").replace("\\", "_").replace(".safetensors", "").replace(".ckpt", "").replace(".gguf", ""))
metadata_file = os.path.join(model_data_dir, "metadata.json")
if os.path.exists(metadata_file) and not force_refresh:
try:
cached = load_json_from_file(metadata_file)
if cached and cached.get("name"):
# Get file modification time for cache date display
cache_mtime = os.path.getmtime(metadata_file)
cache_date = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime(cache_mtime))
print(f"[ConfigBuilder] ✅ Using cached metadata for model: {model_name} (cached on {cache_date})")
return web.json_response({
"metadata": cached,
"saved_to": metadata_file,
"cached": True,
"cache_date": cache_date
})
except Exception:
pass # Cache miss or corrupt file - fall through to fresh lookup
# Resolve the full path based on model type
model_path = None
if model_type == "gguf":
try:
model_path = folder_paths.get_full_path("unet_gguf", model_name)
except (KeyError, Exception):
pass
if model_path is None:
model_path = folder_paths.get_full_path("diffusion_models", model_name)
elif model_type == "diffusion_model":
model_path = folder_paths.get_full_path("diffusion_models", model_name)
else:
model_path = folder_paths.get_full_path("checkpoints", model_name)
if model_path is None:
return web.json_response({
"error": f"Model file not found: {model_name}"
}, status=404)
# Calculate the hash (expensive - only when not cached)
model_hash = calculate_sha256(model_path)
short_hash = model_hash[:10]
print(f"[ConfigBuilder] 📊 Hash calculated: {short_hash}")
# Fetch metadata from CivitAI
model_info = get_model_version_info(model_hash)
if model_info is None:
return web.json_response({
"error": "No metadata found on CivitAI",
"hash": model_hash,
"short_hash": short_hash
}, status=404)
# Extract relevant information
metadata = {
"name": model_info.get("name", "Unknown"),
"model_name": model_info.get("model", {}).get("name", "Unknown") if isinstance(model_info.get("model"), dict) else "Unknown",
"trained_words": model_info.get("trainedWords", []),
"base_model": model_info.get("baseModel", "Unknown"),
"description": model_info.get("description", ""),
"tags": model_info.get("model", {}).get("tags", []) if isinstance(model_info.get("model"), dict) else [],
"images": [],
"url": f"https://civitai.com/models/{model_info.get('modelId', '')}" if model_info.get("modelId") else "",
"hash": model_hash,
"short_hash": short_hash,
"file_path": model_path,
"stats": model_info.get("stats", {}),
"creator": model_info.get("creator", {}).get("username", "Unknown") if isinstance(model_info.get("creator"), dict) else "Unknown"
}
# Extract images
if "images" in model_info and isinstance(model_info["images"], list):
for img in model_info["images"][:5]:
if isinstance(img, dict) and "url" in img:
metadata["images"].append({
"url": img["url"],
"nsfw": img.get("nsfw", "None"),
"width": img.get("width", 0),
"height": img.get("height", 0)
})
# Save metadata to file
output_dir = folder_paths.get_output_directory()
model_data_dir = os.path.join(output_dir, "benchmarks", "model-data", model_name.replace("/", "_").replace("\\", "_").replace(".safetensors", "").replace(".ckpt", "").replace(".gguf", ""))
os.makedirs(model_data_dir, exist_ok=True)
metadata_file = os.path.join(model_data_dir, "metadata.json")
save_dict_to_json(metadata, metadata_file)
print(f"[ConfigBuilder] ✅ Model metadata saved to: {metadata_file}")
return web.json_response({
"metadata": metadata,
"saved_to": metadata_file
})
except Exception as e:
print(f"[ConfigBuilder] ❌ Error in lookup_model_metadata endpoint: {e}")
import traceback
traceback.print_exc()
return web.json_response({
"error": str(e)
}, status=500)
# Fast count endpoint — returns just file counts per category for change detection
@server.PromptServer.instance.routes.get("/configbuilder/model_counts")
async def get_model_counts_endpoint(request):
"""Return file counts per model category. Very fast — uses folder_paths internal cache."""
try:
counts = {}
for cat in ["checkpoints", "diffusion_models", "text_encoders", "vae", "loras"]:
try:
counts[cat] = len(folder_paths.get_filename_list(cat))
except (KeyError, Exception):
counts[cat] = 0
for cat in ["unet_gguf", "clip_gguf", "upscale_models"]:
try:
counts[cat] = len(folder_paths.get_filename_list(cat))
except (KeyError, Exception):
counts[cat] = 0
return web.json_response(counts)
except Exception as e:
return web.json_response({"error": str(e)}, status=500)
# API endpoint to get all model lists for unified model selector
@server.PromptServer.instance.routes.get("/configbuilder/model_lists")
async def get_model_lists_endpoint(request):
"""Return all model lists for the config builder's unified model selector."""
try:
checkpoints = folder_paths.get_filename_list("checkpoints")
diffusion_models = folder_paths.get_filename_list("diffusion_models")
text_encoders = folder_paths.get_filename_list("text_encoders")
# GGUF lists - may not exist if ComfyUI-GGUF is not installed
try:
unet_gguf = folder_paths.get_filename_list("unet_gguf")
except (KeyError, Exception):
unet_gguf = []
try:
clip_gguf = folder_paths.get_filename_list("clip_gguf")
except (KeyError, Exception):
clip_gguf = []
# CLIPType options (matching ComfyUI's CLIPLoader and DualCLIPLoader)
clip_types = [
"stable_diffusion", "stable_cascade", "sd3", "stable_audio",
"mochi", "ltxv", "pixart", "cosmos", "lumina2", "wan",
"hidream", "chroma", "ace", "flux", "flux2", "krea2"
]
dual_clip_types = [
"sdxl", "sd3", "flux", "flux2", "hunyuan_video", "hidream",
"hunyuan_image", "hunyuan_video_15"
]
# Upscale model list
try:
upscale_models = folder_paths.get_filename_list("upscale_models")
except (KeyError, Exception):
upscale_models = []
# Latent upscale model list (separate folder; used by LatentUpscaleModelLoader,
# e.g. LTX 2.3 spatial upscaler)
try:
latent_upscale_models = folder_paths.get_filename_list("latent_upscale_models")
except (KeyError, Exception):
latent_upscale_models = []
# VAE list
vae_list = folder_paths.get_filename_list("vae")
# Sampler and scheduler lists from ComfyUI core
import comfy.samplers
sampler_names = list(comfy.samplers.KSampler.SAMPLERS)
scheduler_names = list(comfy.samplers.KSampler.SCHEDULERS)
return web.json_response({
"checkpoints": checkpoints,
"diffusion_models": diffusion_models,
"unet_gguf": unet_gguf,
"text_encoders": text_encoders,
"clip_gguf": clip_gguf,
"clip_types": clip_types,
"dual_clip_types": dual_clip_types,
"vae": vae_list,
"upscale_models": upscale_models,
"latent_upscale_models": latent_upscale_models,
"samplers": sampler_names,
"schedulers": scheduler_names
})
except Exception as e:
print(f"[ConfigBuilder] Error in model_lists endpoint: {e}")
import traceback
traceback.print_exc()
return web.json_response({"error": str(e)}, status=500)
# API endpoint to refresh model/lora lists (triggered by "Update Node Definitions")
@server.PromptServer.instance.routes.post("/configbuilder/refresh_models")
async def refresh_models_endpoint(request):
"""
API endpoint to signal frontend to clear its caches.
Called when ComfyUI updates node definitions.
"""
try:
print(f"[ConfigBuilder] 🔄 Refresh signal received - clearing frontend caches")
return web.json_response({
"status": "ok",
"message": "Frontend caches should be cleared"
})
except Exception as e:
print(f"[ConfigBuilder] ❌ Error in refresh_models endpoint: {e}")
import traceback
traceback.print_exc()
return web.json_response({
"error": str(e)
}, status=500)
@server.PromptServer.instance.routes.get("/configbuilder/get_lora_triggers")
async def get_lora_triggers_endpoint(request):
"""Get trigger words for a specific LoRA from loras_tags.json"""
try:
lora_name = request.query.get("lora_name", "")
if not lora_name:
return web.json_response({"error": "Missing lora_name"}, status=400)
json_tags_path = os.path.join(folder_paths.get_output_directory(), "benchmarks/loras_tags.json")
triggers = []
if os.path.exists(json_tags_path):
from .lora_utils import load_json_from_file
lora_tags = load_json_from_file(json_tags_path) or {}
# Try exact match, normalized, and backslash variants
normalized = lora_name.replace("\\", "/")
backslash = lora_name.replace("/", "\\")
triggers = lora_tags.get(lora_name, lora_tags.get(normalized, lora_tags.get(backslash, [])))
return web.json_response({"lora_name": lora_name, "triggers": triggers})
except Exception as e:
return web.json_response({"error": str(e)}, status=500)
@server.PromptServer.instance.routes.post("/configbuilder/save_lora_triggers")
async def save_lora_triggers_endpoint(request):
"""Save edited trigger words for a LoRA to loras_tags.json"""
try:
data = await request.json()
lora_name = data.get("lora_name", "")
triggers = data.get("triggers", [])
if not lora_name:
return web.json_response({"error": "Missing lora_name"}, status=400)
json_tags_path = os.path.join(folder_paths.get_output_directory(), "benchmarks/loras_tags.json")
from .lora_utils import load_json_from_file, save_dict_to_json
lora_tags = {}
if os.path.exists(json_tags_path):
lora_tags = load_json_from_file(json_tags_path) or {}
# Normalize the name for consistent storage
normalized = lora_name.replace("\\", "/")
lora_tags[normalized] = triggers
save_dict_to_json(lora_tags, json_tags_path)
# Clear the trigger word LRU cache so changes take effect immediately
from .trigger_words import clear_trigger_caches
clear_trigger_caches()
print(f"[ConfigBuilder] ✏️ Saved {len(triggers)} trigger words for: {normalized}")
return web.json_response({"status": "saved", "lora_name": normalized, "triggers": triggers})
except Exception as e:
return web.json_response({"error": str(e)}, status=500)
@server.PromptServer.instance.routes.post("/configbuilder/preview")
async def preview_endpoint(request):
"""
Preview endpoint — runs state_to_configs_json on the POSTed builder UI
state and returns the resulting configs_json string. The Builder UI's
JSON Preview panel calls this. Single source of truth: this endpoint and
generate_config() share state_to_configs_json, so they cannot disagree.
"""
try:
body = await request.json()
state = body.get("state")
if not isinstance(state, dict):
return web.json_response(
{"error": "Missing or invalid 'state' object in request body"},
status=400,
)
configs_json = UltimateConfigBuilder.state_to_configs_json(state)
return web.json_response({"configs_json": configs_json})
except Exception as e:
return web.json_response({"error": str(e)}, status=500)