# -*- coding: utf-8 -*- """ ComfyUI-H3-Music-Video-AutoShot ================================ H3 Studio (ComfyUI-CGlide / CSGlideCastCS) icin, ComfyUI arayuzunun ICINDEN calisan, sarkiyi beat'lerine gore otomatik sahnelere bolen node. Kullanim: LoadAudio -> H3 Music Video AutoShot -> (Queue Prompt) Node calistiginda: - Sarkiyi beat/tempo analiziyle en fazla max_shot_seconds uzunlugunda, beat sinirlarina yapisik parcalara boler. - Her parcanin GERCEK ses dilimini ComfyUI/input/cglide/ altina wav olarak yazar. - Kaynak H3 Studio workflow JSON'unu okuyup CSGlideCastCS node'unun gcast_project.shots listesini bu sahnelerle DEGISTIRIR (audio referansi + dogru frame uzunlugu + "bu sesi birebir kullan" prompt sablonuyla) ve yeni bir workflow JSON dosyasi olarak yazar. Not: H3 Studio'nun proje verisi o node'un kendi JS arayuzunde tutuluyor, calisan bir ComfyUI grafiginden canli olarak o arayuze yazilamiyor. Bu yuzden bu node CIKTI OLARAK YENI BIR WORKFLOW DOSYASI uretir; sen bu dosyayi ComfyUI'da (File > Open) actiginda sahneler hazir gelir. """ import json import os import random import string import numpy as np FPS = 24 def _make_shot_id(): return "".join(random.choices(string.ascii_lowercase + string.digits, k=7)) def _build_prompt(shot_index, duration_s): return ( f"[SAHNE {shot_index} - buraya gorsel/sahne aciklamasini yaz: " f"karakter(ler), mekan, aksiyon, kamera hareketi]\n\n" f"@audio1 bu sahnenin SES KAYNAGIDIR ve KESIN HUKUMDUR: bu sesi/muzigi " f"OLDUGU GIBI, degistirmeden, yeniden yorumlamadan kullan. Kendi muzigini " f"veya farkli bir ses uretme. Karakterlerin dudak hareketlerini, " f"kostumleri ve varsa dansi/hareketini bu sesin ritmine ve suresine " f"({duration_s:.2f} saniye) birebir senkronla. Sahne, sesin bittigi anda " f"bitmelidir." ) def _detect_segments(y, sr, max_shot_seconds, min_shot_seconds): import librosa duration = float(len(y)) / float(sr) tempo, beat_frames = librosa.beat.beat_track(y=y, sr=sr) try: tempo_val = float(tempo) except TypeError: tempo_val = float(tempo[0]) if len(tempo) else 0.0 beat_times = librosa.frames_to_time(beat_frames, sr=sr).tolist() candidates = sorted(set([0.0] + [round(t, 3) for t in beat_times] + [round(duration, 3)])) segments = [] seg_start = 0.0 while seg_start < duration - 0.05: best = None for t in candidates: if t <= seg_start + min_shot_seconds: continue if t - seg_start <= max_shot_seconds: best = t else: break if best is None: best = min(seg_start + max_shot_seconds, duration) segments.append((seg_start, best)) seg_start = best return tempo_val, segments class H3MusicVideoAutoShot: """Sarkiyi beat'lerine gore H3 Studio sahnelerine otomatik boler.""" CATEGORY = "CGlide/MusicVideo" RETURN_TYPES = ("STRING",) RETURN_NAMES = ("report",) FUNCTION = "run" OUTPUT_NODE = True @classmethod def INPUT_TYPES(cls): return { "required": { "audio": ("AUDIO",), "workflow_path": ("STRING", { "default": r"D:\Comfy\App\ComfyUI_windows_portable\ComfyUI\user\default\workflows\MiniMax_H3_00125_preview.json", "multiline": False, }), "out_path": ("STRING", { "default": r"D:\Comfy\App\ComfyUI_windows_portable\ComfyUI\user\default\workflows\MiniMax_H3_music_video_auto.json", "multiline": False, }), "comfyui_root": ("STRING", { "default": r"D:\Comfy\App\ComfyUI_windows_portable\ComfyUI", "multiline": False, }), "max_shot_seconds": ("FLOAT", {"default": 8.0, "min": 1.0, "max": 15.0, "step": 0.5}), "min_shot_seconds": ("FLOAT", {"default": 2.0, "min": 0.5, "max": 10.0, "step": 0.5}), } } def run(self, audio, workflow_path, out_path, comfyui_root, max_shot_seconds, min_shot_seconds): waveform = audio["waveform"] # [batch, channels, samples] sr = int(audio["sample_rate"]) wav = waveform[0].detach().cpu().numpy() if wav.ndim > 1: y = wav.mean(axis=0) else: y = wav y = y.astype(np.float32) if not os.path.isfile(workflow_path): msg = f"HATA: workflow bulunamadi: {workflow_path}" return (msg,) if not os.path.isdir(comfyui_root): msg = f"HATA: comfyui_root bulunamadi: {comfyui_root}" return (msg,) audio_out_dir = os.path.join(comfyui_root, "input", "cglide") os.makedirs(audio_out_dir, exist_ok=True) with open(workflow_path, "r", encoding="utf-8") as f: wf = json.load(f) cast_node = None for node in wf.get("nodes", []): if node.get("type") == "CSGlideCastCS": cast_node = node break if cast_node is None: return ("HATA: workflow icinde CSGlideCastCS (H3 Studio) node'u bulunamadi.",) project = cast_node["properties"].get("gcast_project") if not project or not project.get("shots"): return ("HATA: gcast_project.shots bos.",) template_shot = project["shots"][0] template_state = template_shot["state"] width = template_state.get("width", 640) height = template_state.get("height", 960) template_images = template_state.get("slots", {}).get("images", [{}] * 9) tempo, segments = _detect_segments(y, sr, max_shot_seconds, min_shot_seconds) import soundfile as sf new_shots = [] lines = [f"Tempo ~{tempo:.1f} BPM, {len(segments)} sahne:"] for i, (start_s, end_s) in enumerate(segments, start=1): dur = end_s - start_s frames = max(1, round(dur * FPS)) audio_filename = f"song_shot_{i:02d}.wav" audio_out_path = os.path.join(audio_out_dir, audio_filename) i0 = int(start_s * sr) i1 = int(end_s * sr) sf.write(audio_out_path, y[i0:i1], sr) shot = { "id": _make_shot_id(), "name": f"Sahne {i}", "off": False, "state": { "mode": "ref2va", "width": width, "height": height, "length": frames, "ref_image_size": "max", "pace": template_state.get("pace", 2.5), "prompt": _build_prompt(i, dur), "slots": { "first": {}, "last": {}, "images": template_images, "videos": [{}, {}, {}], "audios": [{"file": f"cglide/{audio_filename}"}, {}, {}], }, "cont": {}, }, } new_shots.append(shot) lines.append(f" #{i}: {start_s:.2f}s-{end_s:.2f}s ({dur:.2f}s, {frames} frame) -> {audio_filename}") project["shots"] = new_shots project["idx"] = 0 cast_node["properties"]["gcast_project"] = project cast_node["properties"].pop("gcast_project_prev", None) if "widgets_values" in cast_node and cast_node["widgets_values"]: cast_node["widgets_values"][0] = json.dumps(new_shots[0]["state"], ensure_ascii=False) if "widgets_values_named" in cast_node and "h3_data" in cast_node.get("widgets_values_named", {}): cast_node["widgets_values_named"]["h3_data"] = json.dumps(new_shots[0]["state"], ensure_ascii=False) os.makedirs(os.path.dirname(out_path), exist_ok=True) with open(out_path, "w", encoding="utf-8") as f: json.dump(wf, f, ensure_ascii=False) lines.append(f"Yazildi: {out_path}") lines.append("Bu dosyayi ComfyUI'da File > Open ile ac; sahneler hazir gelir.") report = "\n".join(lines) print("[H3MusicVideoAutoShot]\n" + report) return (report,) NODE_CLASS_MAPPINGS = { "H3MusicVideoAutoShot": H3MusicVideoAutoShot, } NODE_DISPLAY_NAME_MAPPINGS = { "H3MusicVideoAutoShot": "H3 Music Video AutoShot", }