diff --git a/__init__.py b/__init__.py index e29d660..a8a2706 100644 --- a/__init__.py +++ b/__init__.py @@ -29,6 +29,10 @@ from .nodes.lora_resize import ( LoRAResizeFrobenius, LoRAResizeCumulative, LoRAMergeToModel ) +from .nodes.downloader_nodes import ( + CheckpointDownloader, LoRADownloader, EmbeddingDownloader, + VAEDownloader, ControlNetDownloader, ManualPathDownloader +) class ModelUtilsExtension(ComfyExtension): @@ -60,6 +64,9 @@ class ModelUtilsExtension(ComfyExtension): LoRAMultiMerge, LoRAMultiMergeDARE, # LoRA Merge To Model LoRAMergeToModel, + # Downloaders + CheckpointDownloader, LoRADownloader, EmbeddingDownloader, + VAEDownloader, ControlNetDownloader, ManualPathDownloader, ] diff --git a/nodes/downloader_nodes.py b/nodes/downloader_nodes.py new file mode 100644 index 0000000..25a36d6 --- /dev/null +++ b/nodes/downloader_nodes.py @@ -0,0 +1,83 @@ +import folder_paths +from comfy_api.latest import io +from .downloader_utils import scan_and_process + +class BaseDownloaderNode(io.ComfyNode): + CATEGORY = "ModelUtils/Downloader" + OUTPUT_NODE = True + + @classmethod + def get_common_inputs(cls): + return { + "recursive": io.BOOLEAN(default=True), + "nsfw_level": io.COMBO(["None", "Soft", "Mature", "X", "XXX", "All"], default="All"), + "max_examples": io.INT(default=0, min=0, max=100), + "threads": io.INT(default=4, min=1, max=32), + "api_key": io.STRING(default="") + } + + @classmethod + def INPUT_TYPES(cls): + return {"required": cls.get_common_inputs()} + + RETURN_TYPES = ("STRING",) + RETURN_NAMES = ("status",) + FUNCTION = "process" + + def process(self, recursive, nsfw_level, max_examples, threads, api_key, **kwargs): + scan_dirs = self.get_scan_dirs(**kwargs) + if not scan_dirs: + return ("No directories to scan.",) + + # Call the scan_and_process function + scan_and_process( + scan_dirs=scan_dirs, + recursive=recursive, + nsfw_level=nsfw_level, + max_examples=max_examples, + api_key=api_key if api_key else None, + threads=threads + ) + + return (f"Completed scanning and downloading for {len(scan_dirs)} directories.",) + + def get_scan_dirs(self, **kwargs) -> list[str]: + return [] + + +class CheckpointDownloader(BaseDownloaderNode): + def get_scan_dirs(self, **kwargs): + return folder_paths.get_folder_paths("checkpoints") + + +class LoRADownloader(BaseDownloaderNode): + def get_scan_dirs(self, **kwargs): + return folder_paths.get_folder_paths("loras") + + +class EmbeddingDownloader(BaseDownloaderNode): + def get_scan_dirs(self, **kwargs): + return folder_paths.get_folder_paths("embeddings") + + +class VAEDownloader(BaseDownloaderNode): + def get_scan_dirs(self, **kwargs): + return folder_paths.get_folder_paths("vae") + + +class ControlNetDownloader(BaseDownloaderNode): + def get_scan_dirs(self, **kwargs): + return folder_paths.get_folder_paths("controlnet") + + +class ManualPathDownloader(BaseDownloaderNode): + @classmethod + def INPUT_TYPES(cls): + inputs = cls.get_common_inputs() + inputs["scan_dir"] = io.STRING(default="") + return {"required": inputs} + + def get_scan_dirs(self, scan_dir, **kwargs): + if scan_dir.strip(): + return [scan_dir.strip()] + return [] diff --git a/nodes/downloader_utils.py b/nodes/downloader_utils.py new file mode 100644 index 0000000..3ae40dd --- /dev/null +++ b/nodes/downloader_utils.py @@ -0,0 +1,567 @@ +import os +import sys +import hashlib +import json +import sqlite3 +import requests +import glob +from pathlib import Path +import threading +import concurrent.futures +import codecs +import subprocess +import shutil + +try: + from PIL import Image + HAS_PIL = True +except ImportError: + HAS_PIL = False + print("Warning: Pillow not found. Image workflow extraction disabled.") + +try: + from mutagen.mp4 import MP4 + HAS_MUTAGEN = True +except ImportError: + HAS_MUTAGEN = False + print("Warning: mutagen not found. MP4 tag extraction will rely entirely on ffmpeg fallback.") + +# Add the reference module to the Python path +# Assuming downloader_utils.py is inside nodes/, the root is one level up +root_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__))) +sys.path.append(os.path.join(root_dir, 'unifiedefficientloader-reference')) +try: + from unifiedefficientloader import UnifiedSafetensorsLoader + HAS_UNIFIED_LOADER = True +except ImportError: + HAS_UNIFIED_LOADER = False + print("Warning: unifiedefficientloader not found. Safetensors metadata extraction disabled.") + +# --- Database Path Logic --- +try: + import folder_paths + base_user_dir = folder_paths.get_public_user_directory("default") + if not base_user_dir: + base_user_dir = root_dir +except (ImportError, AttributeError): + base_user_dir = root_dir + +PLUGIN_DATA_DIR = os.path.join(base_user_dir, "ComfyUI-ModelUtils") +os.makedirs(PLUGIN_DATA_DIR, exist_ok=True) +DEFAULT_DB_PATH = os.path.join(PLUGIN_DATA_DIR, "cache.db") + + +# --- Workflow Extraction Logic --- + +def unescape_and_parse_nested_json(json_string): + if not isinstance(json_string, str): return None + stripped = json_string.strip() + if not (stripped.startswith('{') or stripped.startswith('[')): return None + try: return json.loads(stripped) + except json.JSONDecodeError: pass + try: + first_parse = json.loads(stripped) + if isinstance(first_parse, str): return json.loads(first_parse) + except json.JSONDecodeError: pass + try: + decoded = codecs.decode(stripped, "unicode_escape") + return json.loads(decoded) + except (json.JSONDecodeError, UnicodeDecodeError, ValueError): pass + try: + unescaped = stripped.replace('\\\\"', '\\"').replace('\\"', '"').replace("\\\\", "\\") + unescaped = unescaped.replace("\\n", "\n").replace("\\r", "\r").replace("\\t", "\t") + return json.loads(unescaped) + except json.JSONDecodeError: pass + return None + +def get_workflow_blob(metadata): + for key in ['workflow', 'prompt']: + val = metadata.get(key) + if val: + parsed = unescape_and_parse_nested_json(val) + if parsed: return parsed + for val in metadata.values(): + parsed = unescape_and_parse_nested_json(val) + if isinstance(parsed, dict) and (any(k.isdigit() for k in parsed.keys()) or 'nodes' in parsed): + return parsed + return None + +def process_image_workflow_and_compress(filepath_str): + if not HAS_PIL: return + p = Path(filepath_str) + if p.suffix.lower() not in ['.png', '.jpg', '.jpeg']: return + + workflow = None + actual_format = None + try: + with Image.open(p) as img: + actual_format = img.format + # CivitAI sometimes serves PNGs with .jpeg extension in URL + if actual_format == 'PNG': + workflow = get_workflow_blob(img.info) + if workflow: + # Save workflow + json_path = p.with_suffix('.json') + if not json_path.exists(): + with open(json_path, 'w', encoding='utf-8') as f: + json.dump(workflow, f, indent=2, ensure_ascii=False) + print(f"Extracted workflow to {json_path}") + + # Compress image to WebP + webp_path = p.with_suffix('.webp') + if not webp_path.exists(): + img.save(webp_path, 'WEBP', quality=85, method=4) + print(f"Compressed {p.name} (was actually {actual_format}) to {webp_path.name}") + + # Delete original if we created a webp replacement + if workflow and p.with_suffix('.webp').exists() and actual_format == 'PNG': + os.remove(p) + print(f"Deleted original uncompressed image: {p.name}") + except Exception as e: + print(f"Failed to process image {p.name}: {e}") + +# --- Video Workflow Logic --- +def extract_metadata_ffmpeg(mp4_path): + if not shutil.which('ffmpeg'): return None + try: + cmd = ['ffmpeg', '-i', str(mp4_path), '-f', 'ffmetadata', '-'] + result = subprocess.run(cmd, capture_output=True, text=True, encoding='utf-8', errors='replace') + if result.returncode != 0: return None + metadata, current_key, current_value = {}, None, [] + for line in result.stdout.splitlines(): + if line.startswith(';FFMETADATA'): continue + if '=' in line: + if current_key: metadata[current_key] = '\n'.join(current_value) + parts = line.split('=', 1) + key_lower = parts[0].lower() + if key_lower in ['workflow', 'prompt', 'comment']: + current_key, current_value = key_lower, [parts[1]] + else: current_key = None + elif current_key: + current_value.append(line) + if current_key: metadata[current_key] = '\n'.join(current_value) + return {k: v.replace(r'\=', '=').replace(r'\;', ';').replace(r'\#', '#').replace(r'\\', '\\') for k, v in metadata.items()} + except Exception as e: + print(f"ffmpeg extraction error for {mp4_path}: {e}") + return None + +def get_tag_value(mp4, keys): + for key in keys: + if key in mp4.tags: return mp4.tags[key][0] if mp4.tags[key] else None + for key in keys: + search_suffix = key.split(':')[-1].lower() + for tag_key, val in mp4.tags.items(): + if tag_key.lower().endswith(f":{search_suffix}"): return val[0] if val else None + return None + +def process_video_workflow(filepath_str): + p = Path(filepath_str) + if p.suffix.lower() != '.mp4': return + + json_path = p.with_suffix('.json') + if json_path.exists(): return + + workflow_data = None + mp4 = None + if HAS_MUTAGEN: + try: mp4 = MP4(p) + except: pass + + if mp4: + for tag_keys in [['----:QuickTime:Workflow', 'workflow'], ['----:QuickTime:Prompt', 'prompt'], ['©cmt']]: + val = get_tag_value(mp4, tag_keys) + if val: + if isinstance(val, bytes): val = val.decode('utf-8', errors='ignore') + try: + parsed = json.loads(val) + if isinstance(parsed, dict) and ('prompt' in parsed or any(k.isdigit() for k in parsed.keys())): + if 'prompt' in parsed and isinstance(parsed['prompt'], str): + try: workflow_data = json.loads(parsed['prompt']) + except: workflow_data = parsed + else: + workflow_data = parsed + break + except json.JSONDecodeError: pass + + if not workflow_data: + ffmpeg_data = extract_metadata_ffmpeg(p) + if ffmpeg_data: + for key in ['workflow', 'prompt', 'comment']: + if key in ffmpeg_data: + try: + parsed = json.loads(ffmpeg_data[key]) + if isinstance(parsed, dict): + if 'prompt' in parsed and isinstance(parsed['prompt'], str): + try: workflow_data = json.loads(parsed['prompt']) + except: workflow_data = parsed + else: + workflow_data = parsed + break + except json.JSONDecodeError: pass + + if workflow_data: + try: + with open(json_path, 'w', encoding='utf-8') as f: + json.dump(workflow_data, f, indent=2, ensure_ascii=False) + print(f"Extracted workflow to {json_path.name}") + except Exception as e: + print(f"Error saving workflow for {p.name}: {e}") + +# --- Constants for Civitai NSFW levels --- +NSFW_LEVELS = { + "None": 1, + "Soft": 2, + "Mature": 4, + "X": 8, + "XXX": 16, + "All": 32 # Anything +} + +class HashCache: + def __init__(self, db_path=None): + self.db_path = db_path if db_path else DEFAULT_DB_PATH + self.is_sqlite = self.db_path.endswith('.db') or self.db_path.endswith('.sqlite') + self.lock = threading.Lock() + if self.is_sqlite: + self._init_sqlite() + else: + self.cache = self._load_json() + + def _init_sqlite(self): + with sqlite3.connect(self.db_path) as conn: + cursor = conn.cursor() + cursor.execute(''' + CREATE TABLE IF NOT EXISTS file_hashes ( + filepath TEXT PRIMARY KEY, + mtime REAL, + size INTEGER, + hash TEXT + ) + ''') + conn.commit() + + def _load_json(self): + if os.path.exists(self.db_path): + with open(self.db_path, "r", encoding="utf-8") as f: + return json.load(f) + return {} + + def _save_json(self): + with open(self.db_path, "w", encoding="utf-8") as f: + json.dump(self.cache, f, indent=4) + + def _get_sqlite_hash(self, filepath, mtime, size): + with sqlite3.connect(self.db_path) as conn: + cursor = conn.cursor() + cursor.execute("SELECT mtime, size, hash FROM file_hashes WHERE filepath=?", (filepath,)) + row = cursor.fetchone() + if row: + db_mtime, db_size, db_hash = row + if db_mtime == mtime and db_size == size: + return db_hash + return None + + def _set_sqlite_hash(self, filepath, mtime, size, file_hash): + with sqlite3.connect(self.db_path, timeout=10.0) as conn: + cursor = conn.cursor() + cursor.execute(''' + INSERT OR REPLACE INTO file_hashes (filepath, mtime, size, hash) + VALUES (?, ?, ?, ?) + ''', (filepath, mtime, size, file_hash)) + conn.commit() + + def get_hash(self, filepath): + if not os.path.exists(filepath): + return None + mtime = os.path.getmtime(filepath) + size = os.path.getsize(filepath) + key = filepath + + with self.lock: + if self.is_sqlite: + cached_hash = self._get_sqlite_hash(key, mtime, size) + if cached_hash: return cached_hash + else: + if key in self.cache: + entry = self.cache[key] + if entry["mtime"] == mtime and entry["size"] == size: + return entry["hash"] + + # Calculate new hash (outside the lock to allow parallel hashing) + print(f"Calculating SHA256 for: {filepath}") + sha256_hash = self.calculate_sha256(filepath) + + with self.lock: + if self.is_sqlite: + self._set_sqlite_hash(key, mtime, size, sha256_hash) + else: + self.cache[key] = { + "mtime": mtime, + "size": size, + "hash": sha256_hash + } + self._save_json() + + return sha256_hash + + @staticmethod + def calculate_sha256(filename): + sha256_hash = hashlib.sha256() + with open(filename, "rb") as f: + # Read and update hash string value in blocks of 4K + for byte_block in iter(lambda: f.read(4096), b""): + sha256_hash.update(byte_block) + return sha256_hash.hexdigest() + +class CivitaiAPI: + def __init__(self, api_key=None): + self.api_key = api_key + self.base_url = "https://civitai.com/api/v1" + + def get_headers(self): + headers = {} + if self.api_key: + headers["Authorization"] = f"Bearer {self.api_key}" + return headers + + def get_model_info_by_hash(self, file_hash): + url = f"{self.base_url}/model-versions/by-hash/{file_hash}" + response = requests.get(url, headers=self.get_headers()) + if response.status_code == 200: + return response.json() + elif response.status_code == 404: + return None + else: + response.raise_for_status() + +class Downloader: + @staticmethod + def download_file(url, target_path, headers=None): + if headers is None: + headers = {} + + temp_path = target_path + ".downloading" + + downloaded_size = 0 + if os.path.exists(temp_path): + downloaded_size = os.path.getsize(temp_path) + + if downloaded_size > 0: + headers["Range"] = f"bytes={downloaded_size}-" + + response = requests.get(url, headers=headers, stream=True) + + if response.status_code == 416: # Range not satisfiable, file probably changed or completed + os.remove(temp_path) + downloaded_size = 0 + if "Range" in headers: + del headers["Range"] + response = requests.get(url, headers=headers, stream=True) + + if response.status_code not in [200, 206]: + print(f"Failed to download from {url}, status code: {response.status_code}") + return False + + mode = "ab" if response.status_code == 206 else "wb" + + total_size = int(response.headers.get('content-length', 0)) + downloaded_size + + print(f"Downloading: {target_path} ({downloaded_size}/{total_size} bytes)") + + with open(temp_path, mode) as f: + for chunk in response.iter_content(chunk_size=8192): + if chunk: + f.write(chunk) + + os.rename(temp_path, target_path) + return True + +def get_potential_preview_files(base_path): + exts = ["png", "jpg", "jpeg", "webp", "mp4"] + files = [] + for ext in exts: + files.append(f"{base_path}.preview.{ext}") + files.append(f"{base_path}.{ext}") + return files + +def preview_exists(base_path): + previews = get_potential_preview_files(base_path) + for p in previews: + if os.path.exists(p): + return True + return False + +def next_example_image_path(base_path, ext): + i = 0 + while len(glob.glob(f"{base_path}.example.{i}.*")) > 0: + i += 1 + return f"{base_path}.example.{i}.{ext}" + +def filter_images_by_nsfw(images, threshold_level): + valid_images = [] + for img in images: + rating = img.get("nsfwLevel", 1) + if rating <= threshold_level: + valid_images.append(img) + return valid_images + +def extract_safetensors_metadata(filepath): + if not HAS_UNIFIED_LOADER: + return None + try: + # Use low_memory=True to only read the header and metadata without loading tensors + loader = UnifiedSafetensorsLoader(filepath, low_memory=True) + metadata = loader.metadata() + loader.close() + return metadata if metadata else None + except Exception as e: + print(f"Failed to read safetensors metadata from {filepath} using unifiedefficientloader: {e}") + return None + +def process_file(filepath, cache, api, nsfw_threshold, max_examples): + filepath_str = str(filepath) + base_path = os.path.splitext(filepath_str)[0] + is_safetensors = filepath_str.lower().endswith(".safetensors") + + # Check if all targets already exist + info_path = f"{base_path}.civitai.info" + st_metadata_path = f"{base_path}.metadata.json" + + has_info = os.path.exists(info_path) + has_preview = preview_exists(base_path) + has_st_metadata = not is_safetensors or os.path.exists(st_metadata_path) + + # Process existing images/videos for workflow extraction & compression BEFORE we possibly skip the file + for p in get_potential_preview_files(base_path): + if os.path.exists(p): + ext_lower = p.lower() + if ext_lower.endswith(('.png', '.jpg', '.jpeg')): process_image_workflow_and_compress(p) + elif ext_lower.endswith('.mp4'): process_video_workflow(p) + + for ex in glob.glob(f"{base_path}.example.*"): + ext_lower = ex.lower() + if ext_lower.endswith(('.png', '.jpg', '.jpeg')): process_image_workflow_and_compress(ex) + elif ext_lower.endswith('.mp4'): process_video_workflow(ex) + + # If everything exists and we don't want examples or already have enough examples, we might skip API logic + if has_info and has_preview and has_st_metadata and max_examples == 0: + print(f"Skipping fully populated model: {filepath_str}") + return + + print(f"Processing model: {filepath_str}") + + # Safetensors built-in metadata extraction + if is_safetensors and not has_st_metadata: + metadata = extract_safetensors_metadata(filepath_str) + if metadata: + with open(st_metadata_path, "w", encoding="utf-8") as f: + json.dump(metadata, f, indent=4) + print(f"Saved safetensors metadata to {st_metadata_path}") + + file_hash = cache.get_hash(filepath_str) + if not file_hash: + return + + try: + model_info = api.get_model_info_by_hash(file_hash) + except Exception as e: + print(f"API Error for {filepath_str}: {e}") + return + + if not model_info: + print(f"Model not found on Civitai: {filepath_str}") + return + + # 1. Save Info + if not has_info: + with open(info_path, "w", encoding="utf-8") as f: + json.dump(model_info, f, indent=4) + print(f"Saved metadata to {info_path}") + + images = model_info.get("images", []) + valid_images = filter_images_by_nsfw(images, nsfw_threshold) + + if not valid_images: + print(f"No valid images found matching NSFW criteria for {filepath_str}") + return + + # 2. Save Preview + if not has_preview and len(valid_images) > 0: + preview_img = valid_images[0] + url = preview_img.get("url") + ext = "jpeg" # Default fallback + if ".png" in url.lower(): ext = "png" + elif ".mp4" in url.lower(): ext = "mp4" + elif ".jpg" in url.lower() or ".jpeg" in url.lower(): ext = "jpeg" + + preview_path = f"{base_path}.preview.{ext}" + if Downloader.download_file(url, preview_path): + if ext in ("png", "jpg", "jpeg"): process_image_workflow_and_compress(preview_path) + elif ext == "mp4": process_video_workflow(preview_path) + + # 3. Save Examples + if max_examples > 0: + existing_examples_count = len(glob.glob(f"{base_path}.example.*")) + # We divide by 2 roughly if there are json files, let's just count unique example indices + existing_indices = set() + for f in glob.glob(f"{base_path}.example.*"): + parts = f.split('.example.') + if len(parts) > 1: + idx = parts[1].split('.')[0] + if idx.isdigit(): existing_indices.add(idx) + + needed_examples = max_examples - len(existing_indices) + + if needed_examples > 0: + example_images = valid_images[1:max_examples+1] + # Skip images we probably already downloaded by skipping the first N + example_images_to_dl = example_images[len(existing_indices):] + + for img in example_images_to_dl: + url = img.get("url") + ext = "jpeg" + if ".png" in url.lower(): ext = "png" + elif ".mp4" in url.lower(): ext = "mp4" + elif ".jpg" in url.lower() or ".jpeg" in url.lower(): ext = "jpeg" + + example_path = next_example_image_path(base_path, ext) + if Downloader.download_file(url, example_path): + if ext in ("png", "jpg", "jpeg"): process_image_workflow_and_compress(example_path) + elif ext == "mp4": process_video_workflow(example_path) + +def scan_and_process(scan_dirs, recursive=True, db_path=None, nsfw_level="All", max_examples=0, api_key=None, threads=4): + """ + Scans directories for models and fetches their metadata/previews. + Adapted for direct Python API calls rather than argparse CLI. + """ + db_path = db_path if db_path else DEFAULT_DB_PATH + cache = HashCache(db_path) + api = CivitaiAPI(api_key) + + model_exts = [".safetensors", ".ckpt", ".pt"] + + files_to_process = [] + if isinstance(scan_dirs, str): + scan_dirs = [scan_dirs] + + for scan_dir in scan_dirs: + path = Path(scan_dir) + if recursive: + for ext in model_exts: + files_to_process.extend(path.rglob(f"*{ext}")) + else: + for ext in model_exts: + files_to_process.extend(path.glob(f"*{ext}")) + + nsfw_threshold = NSFW_LEVELS.get(nsfw_level, NSFW_LEVELS["All"]) + + print(f"Found {len(files_to_process)} models to process.") + + if threads > 1: + print(f"Starting {threads} threads...") + with concurrent.futures.ThreadPoolExecutor(max_workers=threads) as executor: + futures = [executor.submit(process_file, fp, cache, api, nsfw_threshold, max_examples) for fp in files_to_process] + concurrent.futures.wait(futures) + else: + for filepath in files_to_process: + process_file(filepath, cache, api, nsfw_threshold, max_examples) diff --git a/standalone_downloader.py b/standalone_downloader.py new file mode 100644 index 0000000..c06d965 --- /dev/null +++ b/standalone_downloader.py @@ -0,0 +1,561 @@ +import os +import sys +import argparse +import hashlib +import json +import sqlite3 +import requests +import glob +from pathlib import Path +import threading +import concurrent.futures +import codecs +import subprocess +import shutil + +try: + from PIL import Image + HAS_PIL = True +except ImportError: + HAS_PIL = False + print("Warning: Pillow not found. Image workflow extraction disabled.") + +try: + from mutagen.mp4 import MP4 + HAS_MUTAGEN = True +except ImportError: + HAS_MUTAGEN = False + print("Warning: mutagen not found. MP4 tag extraction will rely entirely on ffmpeg fallback.") + +# Add the reference module to the Python path +sys.path.append(os.path.join(os.path.dirname(__file__), 'unifiedefficientloader-reference')) +try: + from unifiedefficientloader import UnifiedSafetensorsLoader + HAS_UNIFIED_LOADER = True +except ImportError: + HAS_UNIFIED_LOADER = False + print("Warning: unifiedefficientloader not found. Safetensors metadata extraction disabled.") + +# --- Workflow Extraction Logic --- + +def unescape_and_parse_nested_json(json_string): + if not isinstance(json_string, str): return None + stripped = json_string.strip() + if not (stripped.startswith('{') or stripped.startswith('[')): return None + try: return json.loads(stripped) + except json.JSONDecodeError: pass + try: + first_parse = json.loads(stripped) + if isinstance(first_parse, str): return json.loads(first_parse) + except json.JSONDecodeError: pass + try: + decoded = codecs.decode(stripped, "unicode_escape") + return json.loads(decoded) + except (json.JSONDecodeError, UnicodeDecodeError, ValueError): pass + try: + unescaped = stripped.replace('\\\\"', '\\"').replace('\\"', '"').replace("\\\\", "\\") + unescaped = unescaped.replace("\\n", "\n").replace("\\r", "\r").replace("\\t", "\t") + return json.loads(unescaped) + except json.JSONDecodeError: pass + return None + +def get_workflow_blob(metadata): + for key in ['workflow', 'prompt']: + val = metadata.get(key) + if val: + parsed = unescape_and_parse_nested_json(val) + if parsed: return parsed + for val in metadata.values(): + parsed = unescape_and_parse_nested_json(val) + if isinstance(parsed, dict) and (any(k.isdigit() for k in parsed.keys()) or 'nodes' in parsed): + return parsed + return None + +def process_image_workflow_and_compress(filepath_str): + if not HAS_PIL: return + p = Path(filepath_str) + if p.suffix.lower() not in ['.png', '.jpg', '.jpeg']: return + + workflow = None + actual_format = None + try: + with Image.open(p) as img: + actual_format = img.format + # CivitAI sometimes serves PNGs with .jpeg extension in URL + if actual_format == 'PNG': + workflow = get_workflow_blob(img.info) + if workflow: + # Save workflow + json_path = p.with_suffix('.json') + if not json_path.exists(): + with open(json_path, 'w', encoding='utf-8') as f: + json.dump(workflow, f, indent=2, ensure_ascii=False) + print(f"Extracted workflow to {json_path}") + + # Compress image to WebP + webp_path = p.with_suffix('.webp') + if not webp_path.exists(): + img.save(webp_path, 'WEBP', quality=85, method=4) + print(f"Compressed {p.name} (was actually {actual_format}) to {webp_path.name}") + + # Delete original if we created a webp replacement + if workflow and p.with_suffix('.webp').exists() and actual_format == 'PNG': + os.remove(p) + print(f"Deleted original uncompressed image: {p.name}") + except Exception as e: + print(f"Failed to process image {p.name}: {e}") + +# --- Video Workflow Logic --- +def extract_metadata_ffmpeg(mp4_path): + if not shutil.which('ffmpeg'): return None + try: + cmd = ['ffmpeg', '-i', str(mp4_path), '-f', 'ffmetadata', '-'] + result = subprocess.run(cmd, capture_output=True, text=True, encoding='utf-8', errors='replace') + if result.returncode != 0: return None + metadata, current_key, current_value = {}, None, [] + for line in result.stdout.splitlines(): + if line.startswith(';FFMETADATA'): continue + if '=' in line: + if current_key: metadata[current_key] = '\n'.join(current_value) + parts = line.split('=', 1) + key_lower = parts[0].lower() + if key_lower in ['workflow', 'prompt', 'comment']: + current_key, current_value = key_lower, [parts[1]] + else: current_key = None + elif current_key: + current_value.append(line) + if current_key: metadata[current_key] = '\n'.join(current_value) + return {k: v.replace(r'\=', '=').replace(r'\;', ';').replace(r'\#', '#').replace(r'\\', '\\') for k, v in metadata.items()} + except Exception as e: + print(f"ffmpeg extraction error for {mp4_path}: {e}") + return None + +def get_tag_value(mp4, keys): + for key in keys: + if key in mp4.tags: return mp4.tags[key][0] if mp4.tags[key] else None + for key in keys: + search_suffix = key.split(':')[-1].lower() + for tag_key, val in mp4.tags.items(): + if tag_key.lower().endswith(f":{search_suffix}"): return val[0] if val else None + return None + +def process_video_workflow(filepath_str): + p = Path(filepath_str) + if p.suffix.lower() != '.mp4': return + + json_path = p.with_suffix('.json') + if json_path.exists(): return + + workflow_data = None + mp4 = None + if HAS_MUTAGEN: + try: mp4 = MP4(p) + except: pass + + if mp4: + for tag_keys in [['----:QuickTime:Workflow', 'workflow'], ['----:QuickTime:Prompt', 'prompt'], ['©cmt']]: + val = get_tag_value(mp4, tag_keys) + if val: + if isinstance(val, bytes): val = val.decode('utf-8', errors='ignore') + try: + parsed = json.loads(val) + if isinstance(parsed, dict) and ('prompt' in parsed or any(k.isdigit() for k in parsed.keys())): + if 'prompt' in parsed and isinstance(parsed['prompt'], str): + try: workflow_data = json.loads(parsed['prompt']) + except: workflow_data = parsed + else: + workflow_data = parsed + break + except json.JSONDecodeError: pass + + if not workflow_data: + ffmpeg_data = extract_metadata_ffmpeg(p) + if ffmpeg_data: + for key in ['workflow', 'prompt', 'comment']: + if key in ffmpeg_data: + try: + parsed = json.loads(ffmpeg_data[key]) + if isinstance(parsed, dict): + if 'prompt' in parsed and isinstance(parsed['prompt'], str): + try: workflow_data = json.loads(parsed['prompt']) + except: workflow_data = parsed + else: + workflow_data = parsed + break + except json.JSONDecodeError: pass + + if workflow_data: + try: + with open(json_path, 'w', encoding='utf-8') as f: + json.dump(workflow_data, f, indent=2, ensure_ascii=False) + print(f"Extracted workflow to {json_path.name}") + except Exception as e: + print(f"Error saving workflow for {p.name}: {e}") + +# --- Constants for Civitai NSFW levels --- +NSFW_LEVELS = { + "None": 1, + "Soft": 2, + "Mature": 4, + "X": 8, + "XXX": 16, + "All": 32 # Anything +} + +class HashCache: + def __init__(self, db_path): + self.db_path = db_path + self.is_sqlite = db_path.endswith('.db') or db_path.endswith('.sqlite') + self.lock = threading.Lock() + if self.is_sqlite: + self._init_sqlite() + else: + self.cache = self._load_json() + + def _init_sqlite(self): + with sqlite3.connect(self.db_path) as conn: + cursor = conn.cursor() + cursor.execute(''' + CREATE TABLE IF NOT EXISTS file_hashes ( + filepath TEXT PRIMARY KEY, + mtime REAL, + size INTEGER, + hash TEXT + ) + ''') + conn.commit() + + def _load_json(self): + if os.path.exists(self.db_path): + with open(self.db_path, "r", encoding="utf-8") as f: + return json.load(f) + return {} + + def _save_json(self): + with open(self.db_path, "w", encoding="utf-8") as f: + json.dump(self.cache, f, indent=4) + + def _get_sqlite_hash(self, filepath, mtime, size): + with sqlite3.connect(self.db_path) as conn: + cursor = conn.cursor() + cursor.execute("SELECT mtime, size, hash FROM file_hashes WHERE filepath=?", (filepath,)) + row = cursor.fetchone() + if row: + db_mtime, db_size, db_hash = row + if db_mtime == mtime and db_size == size: + return db_hash + return None + + def _set_sqlite_hash(self, filepath, mtime, size, file_hash): + with sqlite3.connect(self.db_path, timeout=10.0) as conn: + cursor = conn.cursor() + cursor.execute(''' + INSERT OR REPLACE INTO file_hashes (filepath, mtime, size, hash) + VALUES (?, ?, ?, ?) + ''', (filepath, mtime, size, file_hash)) + conn.commit() + + def get_hash(self, filepath): + if not os.path.exists(filepath): + return None + mtime = os.path.getmtime(filepath) + size = os.path.getsize(filepath) + key = filepath + + with self.lock: + if self.is_sqlite: + cached_hash = self._get_sqlite_hash(key, mtime, size) + if cached_hash: return cached_hash + else: + if key in self.cache: + entry = self.cache[key] + if entry["mtime"] == mtime and entry["size"] == size: + return entry["hash"] + + # Calculate new hash (outside the lock to allow parallel hashing) + print(f"Calculating SHA256 for: {filepath}") + sha256_hash = self.calculate_sha256(filepath) + + with self.lock: + if self.is_sqlite: + self._set_sqlite_hash(key, mtime, size, sha256_hash) + else: + self.cache[key] = { + "mtime": mtime, + "size": size, + "hash": sha256_hash + } + self._save_json() + + return sha256_hash + + @staticmethod + def calculate_sha256(filename): + sha256_hash = hashlib.sha256() + with open(filename, "rb") as f: + # Read and update hash string value in blocks of 4K + for byte_block in iter(lambda: f.read(4096), b""): + sha256_hash.update(byte_block) + return sha256_hash.hexdigest() + +class CivitaiAPI: + def __init__(self, api_key=None): + self.api_key = api_key + self.base_url = "https://civitai.com/api/v1" + + def get_headers(self): + headers = {} + if self.api_key: + headers["Authorization"] = f"Bearer {self.api_key}" + return headers + + def get_model_info_by_hash(self, file_hash): + url = f"{self.base_url}/model-versions/by-hash/{file_hash}" + response = requests.get(url, headers=self.get_headers()) + if response.status_code == 200: + return response.json() + elif response.status_code == 404: + return None + else: + response.raise_for_status() + +class Downloader: + @staticmethod + def download_file(url, target_path, headers=None): + if headers is None: + headers = {} + + temp_path = target_path + ".downloading" + + downloaded_size = 0 + if os.path.exists(temp_path): + downloaded_size = os.path.getsize(temp_path) + + if downloaded_size > 0: + headers["Range"] = f"bytes={downloaded_size}-" + + response = requests.get(url, headers=headers, stream=True) + + if response.status_code == 416: # Range not satisfiable, file probably changed or completed + os.remove(temp_path) + downloaded_size = 0 + if "Range" in headers: + del headers["Range"] + response = requests.get(url, headers=headers, stream=True) + + if response.status_code not in [200, 206]: + print(f"Failed to download from {url}, status code: {response.status_code}") + return False + + mode = "ab" if response.status_code == 206 else "wb" + + total_size = int(response.headers.get('content-length', 0)) + downloaded_size + + print(f"Downloading: {target_path} ({downloaded_size}/{total_size} bytes)") + + with open(temp_path, mode) as f: + for chunk in response.iter_content(chunk_size=8192): + if chunk: + f.write(chunk) + + os.rename(temp_path, target_path) + return True + +def get_potential_preview_files(base_path): + exts = ["png", "jpg", "jpeg", "webp", "mp4"] + files = [] + for ext in exts: + files.append(f"{base_path}.preview.{ext}") + files.append(f"{base_path}.{ext}") + return files + +def preview_exists(base_path): + previews = get_potential_preview_files(base_path) + for p in previews: + if os.path.exists(p): + return True + return False + +def next_example_image_path(base_path, ext): + i = 0 + while len(glob.glob(f"{base_path}.example.{i}.*")) > 0: + i += 1 + return f"{base_path}.example.{i}.{ext}" + +def filter_images_by_nsfw(images, threshold_level): + valid_images = [] + for img in images: + rating = img.get("nsfwLevel", 1) + if rating <= threshold_level: + valid_images.append(img) + return valid_images + +def extract_safetensors_metadata(filepath): + if not HAS_UNIFIED_LOADER: + return None + try: + # Use low_memory=True to only read the header and metadata without loading tensors + loader = UnifiedSafetensorsLoader(filepath, low_memory=True) + metadata = loader.metadata() + loader.close() + return metadata if metadata else None + except Exception as e: + print(f"Failed to read safetensors metadata from {filepath} using unifiedefficientloader: {e}") + return None + +def process_file(filepath, args, cache, api, nsfw_threshold): + filepath_str = str(filepath) + base_path = os.path.splitext(filepath_str)[0] + is_safetensors = filepath_str.lower().endswith(".safetensors") + + # Check if all targets already exist + info_path = f"{base_path}.civitai.info" + st_metadata_path = f"{base_path}.metadata.json" + + has_info = os.path.exists(info_path) + has_preview = preview_exists(base_path) + has_st_metadata = not is_safetensors or os.path.exists(st_metadata_path) + + # Process existing images/videos for workflow extraction & compression BEFORE we possibly skip the file + for p in get_potential_preview_files(base_path): + if os.path.exists(p): + ext_lower = p.lower() + if ext_lower.endswith(('.png', '.jpg', '.jpeg')): process_image_workflow_and_compress(p) + elif ext_lower.endswith('.mp4'): process_video_workflow(p) + + for ex in glob.glob(f"{base_path}.example.*"): + ext_lower = ex.lower() + if ext_lower.endswith(('.png', '.jpg', '.jpeg')): process_image_workflow_and_compress(ex) + elif ext_lower.endswith('.mp4'): process_video_workflow(ex) + + # If everything exists and we don't want examples or already have enough examples, we might skip API logic + if has_info and has_preview and has_st_metadata and args.max_examples == 0: + print(f"Skipping fully populated model: {filepath_str}") + return + + print(f"Processing model: {filepath_str}") + + # Safetensors built-in metadata extraction + if is_safetensors and not has_st_metadata: + metadata = extract_safetensors_metadata(filepath_str) + if metadata: + with open(st_metadata_path, "w", encoding="utf-8") as f: + json.dump(metadata, f, indent=4) + print(f"Saved safetensors metadata to {st_metadata_path}") + + file_hash = cache.get_hash(filepath_str) + if not file_hash: + return + + try: + model_info = api.get_model_info_by_hash(file_hash) + except Exception as e: + print(f"API Error for {filepath_str}: {e}") + return + + if not model_info: + print(f"Model not found on Civitai: {filepath_str}") + return + + # 1. Save Info + if not has_info: + with open(info_path, "w", encoding="utf-8") as f: + json.dump(model_info, f, indent=4) + print(f"Saved metadata to {info_path}") + + images = model_info.get("images", []) + valid_images = filter_images_by_nsfw(images, nsfw_threshold) + + if not valid_images: + print(f"No valid images found matching NSFW criteria for {filepath_str}") + return + + # 2. Save Preview + if not has_preview and len(valid_images) > 0: + preview_img = valid_images[0] + url = preview_img.get("url") + ext = "jpeg" # Default fallback + if ".png" in url.lower(): ext = "png" + elif ".mp4" in url.lower(): ext = "mp4" + elif ".jpg" in url.lower() or ".jpeg" in url.lower(): ext = "jpeg" + + preview_path = f"{base_path}.preview.{ext}" + if Downloader.download_file(url, preview_path): + if ext in ("png", "jpg", "jpeg"): process_image_workflow_and_compress(preview_path) + elif ext == "mp4": process_video_workflow(preview_path) + + # 3. Save Examples + if args.max_examples > 0: + existing_examples_count = len(glob.glob(f"{base_path}.example.*")) + # We divide by 2 roughly if there are json files, let's just count unique example indices + existing_indices = set() + for f in glob.glob(f"{base_path}.example.*"): + parts = f.split('.example.') + if len(parts) > 1: + idx = parts[1].split('.')[0] + if idx.isdigit(): existing_indices.add(idx) + + needed_examples = args.max_examples - len(existing_indices) + + if needed_examples > 0: + example_images = valid_images[1:args.max_examples+1] + # Skip images we probably already downloaded by skipping the first N + example_images_to_dl = example_images[len(existing_indices):] + + for img in example_images_to_dl: + url = img.get("url") + ext = "jpeg" + if ".png" in url.lower(): ext = "png" + elif ".mp4" in url.lower(): ext = "mp4" + elif ".jpg" in url.lower() or ".jpeg" in url.lower(): ext = "jpeg" + + example_path = next_example_image_path(base_path, ext) + if Downloader.download_file(url, example_path): + if ext in ("png", "jpg", "jpeg"): process_image_workflow_and_compress(example_path) + elif ext == "mp4": process_video_workflow(example_path) + +def scan_and_process(args): + cache = HashCache(args.db_path) + api = CivitaiAPI(args.api_key) + + model_exts = [".safetensors", ".ckpt", ".pt"] + + files_to_process = [] + for scan_dir in args.scan_dirs: + path = Path(scan_dir) + if args.recursive: + for ext in model_exts: + files_to_process.extend(path.rglob(f"*{ext}")) + else: + for ext in model_exts: + files_to_process.extend(path.glob(f"*{ext}")) + + nsfw_threshold = NSFW_LEVELS.get(args.nsfw_level, 1) + + print(f"Found {len(files_to_process)} models to process.") + + if args.threads > 1: + print(f"Starting {args.threads} threads...") + with concurrent.futures.ThreadPoolExecutor(max_workers=args.threads) as executor: + futures = [executor.submit(process_file, fp, args, cache, api, nsfw_threshold) for fp in files_to_process] + concurrent.futures.wait(futures) + else: + for filepath in files_to_process: + process_file(filepath, args, cache, api, nsfw_threshold) + +def main(): + parser = argparse.ArgumentParser(description="Standalone Civitai Model Downloader") + parser.add_argument("--scan-dirs", nargs="+", required=True, help="Directories to scan for models") + parser.add_argument("--recursive", action="store_true", help="Recursively scan directories") + parser.add_argument("--db-path", default="model_hash_cache.json", help="Path to hash cache JSON") + parser.add_argument("--nsfw-level", choices=list(NSFW_LEVELS.keys()), default="Soft", help="Max NSFW level for images") + parser.add_argument("--max-examples", type=int, default=0, help="Max number of example images to download") + parser.add_argument("--max-size-preview", action="store_true", help="Download max resolution previews (Not fully implemented, fallback to default URL)") + parser.add_argument("--api-key", default=None, help="Civitai API Key") + parser.add_argument("--threads", type=int, default=1, help="Number of concurrent threads for hashing and downloading") + + args = parser.parse_args() + scan_and_process(args) + +if __name__ == "__main__": + main() \ No newline at end of file