| import os
|
| import json
|
| import folder_paths
|
| import comfy.sd
|
| import comfy.utils
|
| from safetensors import safe_open
|
| import hashlib
|
| import urllib.request
|
|
|
| CACHE_FILE = os.path.join(os.path.dirname(__file__), "dolphin_tags_cache.json")
|
|
|
| def load_dolphin_cache():
|
| if os.path.exists(CACHE_FILE):
|
| try:
|
| with open(CACHE_FILE, "r", encoding="utf-8") as f:
|
| return json.load(f)
|
| except: return {}
|
| return {}
|
|
|
| def save_dolphin_cache(cache_data):
|
| try:
|
| with open(CACHE_FILE, "w", encoding="utf-8") as f:
|
| json.dump(cache_data, f, indent=4)
|
| except: pass
|
|
|
| class DolphinArchitectV15:
|
| @classmethod
|
| def INPUT_TYPES(s):
|
| loras = ["None"] + folder_paths.get_filename_list("loras")
|
|
|
| res = {
|
| "required": {
|
| "global_switch": ("BOOLEAN", {"default": True, "label_on": "GLOBAL ON", "label_off": "GLOBAL OFF (SKIP)"}),
|
| "model_base": ("MODEL",), "clip_base": ("CLIP",),
|
| "model_high": ("MODEL",), "clip_high": ("CLIP",),
|
| "model_low": ("MODEL",), "clip_low": ("CLIP",),
|
| },
|
| "optional": {}
|
| }
|
|
|
|
|
| res["optional"]["header_master"] = ("STRING", {"default": "โโโ [ MASTER: B+H ] โโโ", "multiline": False})
|
| for i in range(1, 5):
|
| res["optional"][f"master_l_{i}"] = (loras, {"default": "None"})
|
| res["optional"][f"master_b_w_{i}"] = ("FLOAT", {"default": 1.0, "min": -10.0, "max": 10.0, "step": 0.01})
|
| res["optional"][f"master_h_w_{i}"] = ("FLOAT", {"default": 1.0, "min": -10.0, "max": 10.0, "step": 0.01})
|
|
|
| res["optional"]["header_low"] = ("STRING", {"default": "โโโ [ LOW ONLY ] โโโ", "multiline": False})
|
| for i in range(1, 5):
|
| res["optional"][f"low_l_{i}"] = (loras, {"default": "None"})
|
| res["optional"][f"low_w_{i}"] = ("FLOAT", {"default": 1.0, "min": -10.0, "max": 10.0, "step": 0.01})
|
|
|
| res["optional"]["manual_triggers"] = ("STRING", {"default": "", "multiline": True})
|
| return res
|
|
|
| RETURN_TYPES = ("MODEL", "CLIP", "MODEL", "CLIP", "MODEL", "CLIP", "STRING")
|
| RETURN_NAMES = ("M_BASE", "C_BASE", "M_HIGH", "C_HIGH", "M_LOW", "C_LOW", "triggers")
|
| FUNCTION = "exec"
|
| CATEGORY = "Dolphin"
|
|
|
| def get_trig(self, lora_name):
|
| if lora_name == "None": return []
|
| path = folder_paths.get_full_path("loras", lora_name)
|
| if not path or not os.path.exists(path): return []
|
|
|
| triggers = []
|
| try:
|
| with safe_open(path, framework="pt", device="cpu") as f:
|
| m = f.metadata()
|
| if m and "modelspec.trigger_words" in m:
|
| tw = json.loads(m["modelspec.trigger_words"])
|
| if isinstance(tw, list): triggers.extend(tw)
|
| elif isinstance(tw, str): triggers.append(tw)
|
| except: pass
|
| if triggers: return triggers
|
|
|
| cache = load_dolphin_cache()
|
| if lora_name in cache and cache[lora_name]: return cache[lora_name]
|
|
|
| try:
|
| sha256_hash = hashlib.sha256()
|
| with open(path, "rb") as f:
|
| for byte_block in iter(lambda: f.read(4096 * 1024), b""):
|
| sha256_hash.update(byte_block)
|
|
|
| url = f"https://civitai.com/api/v1/model-versions/by-hash/{sha256_hash.hexdigest()}"
|
| req = urllib.request.Request(url, headers={'User-Agent': 'Mozilla/5.0'})
|
| with urllib.request.urlopen(req, timeout=5) as response:
|
| data = json.loads(response.read().decode('utf-8'))
|
| api_triggers = data.get("trainedWords", [])
|
| if api_triggers:
|
| cache[lora_name] = api_triggers
|
| save_dolphin_cache(cache)
|
| return api_triggers
|
| except: pass
|
| return triggers
|
|
|
| def exec(self, global_switch, model_base, clip_base, model_high, clip_high, model_low, clip_low, manual_triggers="", **kwargs):
|
| mb, cb = model_base, clip_base
|
| mh, ch = model_high, clip_high
|
| ml, cl = model_low, clip_low
|
|
|
| if not global_switch:
|
| trigs = [t.strip() for t in manual_triggers.split(',') if t.strip()]
|
| ut = ", ".join(list(dict.fromkeys(trigs)))
|
| return (mb, cb, mh, ch, ml, cl, ut)
|
|
|
| applied_names = []
|
|
|
| for i in range(1, 5):
|
| n = kwargs.get(f"master_l_{i}", "None")
|
| w_b = kwargs.get(f"master_b_w_{i}", 1.0)
|
| w_h = kwargs.get(f"master_h_w_{i}", 1.0)
|
|
|
|
|
| if n != "None" and (w_b != 0.0 or w_h != 0.0):
|
| applied_names.append(n)
|
| lora_data = comfy.utils.load_torch_file(folder_paths.get_full_path("loras", n))
|
| mb, cb = comfy.sd.load_lora_for_models(mb, cb, lora_data, w_b, w_b)
|
| mh, ch = comfy.sd.load_lora_for_models(mh, ch, lora_data, w_h, w_h)
|
|
|
| for i in range(1, 5):
|
| n = kwargs.get(f"low_l_{i}", "None")
|
| w = kwargs.get(f"low_w_{i}", 1.0)
|
|
|
| if n != "None" and w != 0.0:
|
| applied_names.append(n)
|
| lora_data = comfy.utils.load_torch_file(folder_paths.get_full_path("loras", n))
|
| ml, cl = comfy.sd.load_lora_for_models(ml, cl, lora_data, w, w)
|
|
|
| trigs = []
|
| for n in set(applied_names):
|
| trigs.extend(self.get_trig(n))
|
|
|
| if manual_triggers.strip():
|
| manual_list = [t.strip() for t in manual_triggers.split(',') if t.strip()]
|
| trigs.extend(manual_list)
|
|
|
| unique_trigs = ", ".join(list(dict.fromkeys(trigs)))
|
|
|
| return (mb, cb, mh, ch, ml, cl, unique_trigs) |
