""" Visual domain — image generation node'ları + visual reference (IPAdapter/PuLID/ ControlNet) helpers, model availability detection, reference chain. Mevcut Node sınıfları (_legacy.py'dan re-export): - FikoYT_MVD_SegmentImageGen - FikoYT_MVD_HiResFix - FikoYT_MVD_AlbumCoverGen - FikoYT_MVD_ImageAspectBatch Yeni Node sınıfları (Sprint 2 — W13): - FikoYT_MVD_PoseExtractor — DWPose/OpenPose ile keypoint extract - FikoYT_MVD_ControlNetUnion — Tek slot'ta pose+canny+depth+softedge Yeni Node sınıfları (Sprint 3 — W14): - FikoYT_MVD_RegionalMultiChar — 2/3/4 karakter regional prompting Yeni Node sınıfları (Sprint 5 — W16): - FikoYT_MVD_LightningPreview — SDXL Lightning 4-step preview """ from ._legacy import ( FikoYT_MVD_SegmentImageGen, FikoYT_MVD_HiResFix, FikoYT_MVD_AlbumCoverGen, FikoYT_MVD_ImageAspectBatch, _empty_image, _get_model_search_paths, _try_apply_flux_guidance, _try_apply_pulid, _load_input_image, _try_apply_ipadapter, _try_run_preprocessor, _try_apply_controlnet, _classify_ref_mode, _apply_references_chain, _get_missing_flux_models, _download_one_sync, _comfy_class_available, _find_controlnet_file, _visual_ref_capabilities, _detect_pipeline_nodes, _next_free_id, _inject_visual_references, CATEGORY, ) # ============================================================================= # Sprint 2 (W13): ControlNet Union helper # ============================================================================= def _try_apply_controlnet_union(positive_cond, negative_cond, ref_image, modes=("pose",), weights=(0.6,), vae=None, union_model_hint=None): """ControlNet Union — tek slot'ta birden fazla kontrolü birleştir. SDXL: xinsir/controlnet-union-sdxl-1.0 Flux: Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro veya InstantX variant modes: tuple of "pose"|"canny"|"depth"|"softedge" weights: aynı uzunlukta tuple veya tek float (replicate edilir) Her mode için preprocessor sonucu hesaplanıp ControlNet Union'a sırayla uygulanır. Union model bulunamazsa standard ControlNet'e fallback. Dönüş: (new_positive, new_negative, status_str) """ if ref_image is None: return positive_cond, negative_cond, "ControlNet Union: ref yok" if not modes: return positive_cond, negative_cond, "ControlNet Union: mod yok" # weights normalize if isinstance(weights, (int, float)): weights = [float(weights)] * len(modes) weights = list(weights) while len(weights) < len(modes): weights.append(weights[-1] if weights else 0.6) try: from nodes import NODE_CLASS_MAPPINGS as _NCM import folder_paths cn_loader = _NCM.get("ControlNetLoader") cn_apply = (_NCM.get("ControlNetApplyAdvanced") or _NCM.get("ControlNetApplySD3") or _NCM.get("ControlNetApply")) if not cn_loader or not cn_apply: return positive_cond, negative_cond, "ControlNet node'ları yok" # Union model bul try: cn_files = folder_paths.get_filename_list("controlnet") except Exception: cn_files = [] union_file = None priority_terms = ("union",) flux_terms = ("flux", "instantx", "shakker", "xlabs") candidates = [f for f in cn_files if any(t in f.lower() for t in priority_terms)] if union_model_hint: for f in candidates: if union_model_hint.lower() in f.lower(): union_file = f break if not union_file: # Flux Union > SDXL Union > diğer Union flux_union = [f for f in candidates if any(t in f.lower() for t in flux_terms)] union_file = flux_union[0] if flux_union else (candidates[0] if candidates else None) if not union_file: # Standart ControlNet fallback — mode başına ayrı CN status_parts = ["ControlNet Union modeli yok, standard ControlNet'e fallback"] cur_pos, cur_neg = positive_cond, negative_cond for m, w in zip(modes, weights): cur_pos, cur_neg, msg = _try_apply_controlnet( cur_pos, cur_neg, ref_image, f"controlnet_{m}" if m in ("canny", "depth", "normal") else m, weight=w, vae=vae) status_parts.append(f"[{m}] {msg}") return cur_pos, cur_neg, " | ".join(status_parts) # Union model var — her mod için preprocess + apply try: cn_model = cn_loader().load_controlnet(union_file)[0] except Exception as e: return positive_cond, negative_cond, f"ControlNet Union load hata: {e}" cur_pos, cur_neg = positive_cond, negative_cond status_parts = [] for m, w in zip(modes, weights): if w <= 0: continue prep_mode = m if m in ("openpose", "dwpose") else f"controlnet_{m}" ctrl_image, prep_msg = _try_run_preprocessor(ref_image, prep_mode) if ctrl_image is None: status_parts.append(f"[{m}] preprocess fail") continue try: result = cn_apply().apply_controlnet( cur_pos, cur_neg, cn_model, ctrl_image, strength=w, start_percent=0.0, end_percent=1.0, vae=vae ) except TypeError: result = cn_apply().apply_controlnet( cur_pos, cur_neg, cn_model, ctrl_image, strength=w, start_percent=0.0, end_percent=1.0 ) cur_pos, cur_neg = result[0], result[1] status_parts.append(f"✓[{m},w={w}]") return cur_pos, cur_neg, ( f"✓ ControlNet Union ({union_file}): " + " ".join(status_parts)) except Exception as e: import traceback print(traceback.format_exc()[-500:]) return positive_cond, negative_cond, f"ControlNet Union hata: {e}" # ============================================================================= # Sprint 2 (W13): FikoYT_MVD_PoseExtractor node # ============================================================================= class FikoYT_MVD_PoseExtractor: """Bir giriş resminden pose keypoint çıkar (DWPose öncelikli, fallback OpenPose). Çıktı: pose_image (skeleton overlay) + json string (keypoints) + status. comfyui_controlnet_aux kurulu değilse raw image + boş json + uyarı döner. """ CATEGORY = CATEGORY FUNCTION = "extract" RETURN_TYPES = ("IMAGE", "STRING", "STRING") RETURN_NAMES = ("pose_image", "keypoints_json", "status") @classmethod def INPUT_TYPES(cls): return { "required": { "image": ("IMAGE",), "preferred": (["dwpose", "openpose"], {"default": "dwpose"}), "resolution": ("INT", {"default": 512, "min": 256, "max": 2048, "step": 64}), } } def extract(self, image, preferred="dwpose", resolution=512): try: from nodes import NODE_CLASS_MAPPINGS as _NCM order = ["DWPreprocessor", "OpenposePreprocessor"] if preferred == "openpose": order = ["OpenposePreprocessor", "DWPreprocessor"] pose_image = image keypoints = "{}" status = "Preprocessor yok" for node_name in order: cls = _NCM.get(node_name) if not cls: continue try: node = cls() method = None for m in ("execute", "estimate_pose", "process"): if hasattr(node, m): method = getattr(node, m) break if method is None: continue # DWPose: image=, detect_hand=, detect_body=, detect_face=, resolution= try: result = method(image=image, detect_hand="enable", detect_body="enable", detect_face="enable", resolution=resolution) except TypeError: result = method(image=image, resolution=resolution) # Sprint 38: 3 farklı return format desteği # 1. (tensor, json) — eski tuple # 2. tensor — tek # 3. {"ui": {...}, "result": (tensor, json)} — ComfyUI 0.3+ dict import json inner = result if isinstance(result, dict): # ComfyUI yeni format if "result" in result: inner = result["result"] elif "ui" in result and "images" in result.get("ui", {}): # Sadece ui dönmüş, tensor yok — orijinal image kullan inner = (image,) else: inner = (image,) if isinstance(inner, (tuple, list)) and len(inner) > 0: pose_image = inner[0] if len(inner) > 1: try: keypoints = json.dumps(inner[1]) if not isinstance( inner[1], str) else inner[1] except Exception: keypoints = "{}" else: pose_image = inner if inner is not None else image # Sanity: pose_image bir torch tensor olmalı if not hasattr(pose_image, "shape"): print(f"[FikoYT PoseExtractor] {node_name} returned non-tensor: " f"{type(pose_image).__name__}, falling back to input") pose_image = image status = f"✓ {node_name} ({resolution}px)" break except Exception as e: print(f"[FikoYT] PoseExtractor {node_name} hata: {e}") continue return (pose_image, keypoints, status) except Exception as e: return (image, "{}", f"PoseExtractor hata: {e}") # ============================================================================= # Sprint 2 (W13): FikoYT_MVD_ControlNetUnion node # ============================================================================= class FikoYT_MVD_ControlNetUnion: """ControlNet Union node wrapper — tek slot'ta pose+canny+depth+softedge. Dahili olarak _try_apply_controlnet_union helper'ını çağırır. Workflow JSON'da standalone kullanılabilir, MVD render pipeline'ı da bu node'u çağırabilir. """ CATEGORY = CATEGORY FUNCTION = "apply" RETURN_TYPES = ("CONDITIONING", "CONDITIONING", "STRING") RETURN_NAMES = ("positive", "negative", "status") @classmethod def INPUT_TYPES(cls): return { "required": { "positive": ("CONDITIONING",), "negative": ("CONDITIONING",), "control_image": ("IMAGE",), "enable_pose": ("BOOLEAN", {"default": True}), "enable_canny": ("BOOLEAN", {"default": False}), "enable_depth": ("BOOLEAN", {"default": False}), "enable_softedge": ("BOOLEAN", {"default": False}), "weight_pose": ("FLOAT", {"default": 0.6, "min": 0.0, "max": 2.0, "step": 0.05}), "weight_canny": ("FLOAT", {"default": 0.4, "min": 0.0, "max": 2.0, "step": 0.05}), "weight_depth": ("FLOAT", {"default": 0.4, "min": 0.0, "max": 2.0, "step": 0.05}), "weight_softedge": ("FLOAT", {"default": 0.3, "min": 0.0, "max": 2.0, "step": 0.05}), }, "optional": { "vae": ("VAE",), "union_model_hint": ("STRING", {"default": "", "multiline": False}), } } def apply(self, positive, negative, control_image, enable_pose=True, enable_canny=False, enable_depth=False, enable_softedge=False, weight_pose=0.6, weight_canny=0.4, weight_depth=0.4, weight_softedge=0.3, vae=None, union_model_hint=""): modes, weights = [], [] if enable_pose: modes.append("dwpose"); weights.append(weight_pose) if enable_canny: modes.append("canny"); weights.append(weight_canny) if enable_depth: modes.append("depth"); weights.append(weight_depth) if enable_softedge: modes.append("softedge"); weights.append(weight_softedge) if not modes: return (positive, negative, "ControlNet Union: hiçbir mod aktif değil") new_pos, new_neg, status = _try_apply_controlnet_union( positive, negative, control_image, modes=modes, weights=weights, vae=vae, union_model_hint=union_model_hint or None ) return (new_pos, new_neg, status) # ============================================================================= # Sprint 3 (W14): FikoYT_MVD_RegionalMultiChar — 2/3/4 karakter regional # ============================================================================= class FikoYT_MVD_RegionalMultiChar: """Multi-character regional prompting wrapper. Tek code path'te 2/3/4 karakter destekler. characters_json bir JSON string bekler: [{cid, lora_strength, mask_polygon: [[x,y],...], prompt_override?, identity_method?}, ...]. base_prompt arka plan + ortam için. Backend implementasyonu Sprint 3'te tam doluyor (workflow JSON template'leri flux_regional_3char.json/4char.json). Bu node şu an helper stub'ı. """ CATEGORY = CATEGORY FUNCTION = "build" RETURN_TYPES = ("STRING", "STRING") RETURN_NAMES = ("plan_json", "status") @classmethod def INPUT_TYPES(cls): return { "required": { "characters_json": ("STRING", { "default": "[]", "multiline": True, "placeholder": '[{"cid":"singer","lora_strength":0.8,"mask_polygon":[[0,0],[256,0],[256,512],[0,512]]}]' }), "base_prompt": ("STRING", {"default": "cinematic film still", "multiline": True}), "width": ("INT", {"default": 1024, "min": 256, "max": 4096, "step": 64}), "height": ("INT", {"default": 1024, "min": 256, "max": 4096, "step": 64}), } } def build(self, characters_json, base_prompt, width, height): import json as _json try: chars = _json.loads(characters_json or "[]") except Exception as e: return ("{}", f"characters_json parse hata: {e}") if not isinstance(chars, list): return ("{}", "characters_json bir liste olmalı") n = len(chars) if n < 2 or n > 4: return ("{}", f"2-4 karakter destekleniyor, {n} verildi") plan = { "char_count": n, "characters": chars, "base_prompt": base_prompt, "width": width, "height": height, "workflow_template": f"flux_regional_{n}char.json", } return (_json.dumps(plan, ensure_ascii=False, indent=2), f"✓ Regional plan: {n} karakter, template={plan['workflow_template']}") # ============================================================================= # Sprint 5 (W16): FikoYT_MVD_LightningPreview — SDXL Lightning 4-step # ============================================================================= class FikoYT_MVD_LightningPreview: """SDXL Lightning 4-step preview render — hızlı iterasyon. Lightning LoRA modelle birleştirilir (sampler="euler", scheduler="sgm_uniform", steps=4, cfg=1.0). Flux modellerinde no-op (uyarı döner). Workflow JSON adapter olarak da kullanılabilir. """ CATEGORY = CATEGORY FUNCTION = "preview" RETURN_TYPES = ("MODEL", "STRING") RETURN_NAMES = ("model", "status") @classmethod def INPUT_TYPES(cls): return { "required": { "model": ("MODEL",), "lightning_lora": ("STRING", { "default": "sdxl_lightning_4step_lora.safetensors", "multiline": False, "placeholder": "models/loras/ altındaki Lightning LoRA dosya adı" }), "strength": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 2.0, "step": 0.05}), } } def preview(self, model, lightning_lora, strength=1.0): try: from nodes import NODE_CLASS_MAPPINGS as _NCM cls = _NCM.get("LoraLoaderModelOnly") or _NCM.get("LoraLoader") if not cls: return (model, "Lightning: LoraLoader node yok") try: # LoraLoaderModelOnly: (model, lora_name, strength_model) patched = cls().load_lora_model_only(model, lightning_lora, strength)[0] except Exception: # LoraLoader: (model, clip, lora_name, strength_model, strength_clip) # clip yoksa skip return (model, "Lightning: LoraLoaderModelOnly yok, manuel kur") return (patched, f"✓ Lightning LoRA uygulandı ({lightning_lora}, s={strength}) — " f"4 step euler/sgm_uniform önerilir") except Exception as e: return (model, f"Lightning hata: {e}") # ============================================================================= # Sprint 6 (Production Pipeline): Profesyonel post-grade + Master render + Aspect preset # ============================================================================= # Müzik klibi için endüstri-standardı aspect ratio preset'leri # Her preset → (width, height, label, platform) MUSIC_VIDEO_ASPECT_PRESETS = { "tiktok_reels": (1080, 1920, "TikTok / Reels / Shorts (9:16)", "tiktok"), "instagram": (1080, 1080, "Instagram Feed (1:1)", "instagram"), "instagram_vertical": (1080, 1350, "Instagram Vertical (4:5)", "instagram"), "youtube": (1920, 1080, "YouTube / Vimeo (16:9)", "youtube"), "youtube_4k": (3840, 2160, "YouTube 4K (16:9)", "youtube"), "cinematic": (2048, 858, "Sinematik (2.39:1 letterbox)", "film"), "ultrawide": (2560, 1080, "Ultrawide (21:9)", "monitor"), } def _apply_lut_3d(image_tensor, lut_name="cinematic", strength=1.0): """Basit 3D LUT yaklaşımı: birkaç preset için kanal-tabanlı ton eğrisi. Profesyonel kullanım: .cube LUT dosyalarını models/luts/ altına koyup color-matcher lib ile aralarda interpolation yapılır. Burada hafif/preset. lut_name: "cinematic" | "vintage" | "vibrant" | "teal_orange" | "noir" | "warm" | "cool" """ try: import torch import numpy as np except ImportError: return image_tensor, "torch/numpy yok" if strength <= 0: return image_tensor, "LUT: strength=0, atlandı" out = image_tensor.clone() # LUT preset'leri — (R_curve, G_curve, B_curve) shadow/midtone/highlight şiddetleri presets = { # cinematic: kontrast + sıcak shadow + soğuk highlight "cinematic": {"shadow": (0.95, 0.97, 1.08), "midtone": (1.02, 1.0, 1.0), "highlight": (1.05, 1.02, 0.92), "saturation": 0.92, "contrast": 1.08}, # vintage: düşük kontrast, sarı/yeşil tonlama, lifted blacks "vintage": {"shadow": (1.08, 1.02, 0.88), "midtone": (1.05, 1.02, 0.95), "highlight": (1.02, 1.0, 0.85), "saturation": 0.78, "contrast": 0.92, "lift": 0.05}, # vibrant: doygunluk + canlılık (pop music) "vibrant": {"shadow": (1.0, 1.0, 1.0), "midtone": (1.0, 1.0, 1.0), "highlight": (1.0, 1.0, 1.0), "saturation": 1.3, "contrast": 1.12}, # teal & orange: Hollywood blockbuster "teal_orange": {"shadow": (0.92, 1.05, 1.18), "midtone": (1.02, 1.0, 0.95), "highlight": (1.15, 1.05, 0.85), "saturation": 1.05, "contrast": 1.1}, # noir: yüksek kontrast siyah-beyaz tilt "noir": {"shadow": (1.0, 1.0, 1.0), "midtone": (1.0, 1.0, 1.0), "highlight": (1.0, 1.0, 1.0), "saturation": 0.15, "contrast": 1.25}, # warm: sıcak ton (akustik/duygusal) "warm": {"shadow": (1.08, 1.02, 0.92), "midtone": (1.05, 1.0, 0.95), "highlight": (1.08, 1.02, 0.92), "saturation": 1.05, "contrast": 1.02}, # cool: soğuk ton (elektronik/melankolik) "cool": {"shadow": (0.92, 0.98, 1.08), "midtone": (0.95, 1.0, 1.05), "highlight": (0.92, 0.98, 1.08), "saturation": 1.0, "contrast": 1.05}, } p = presets.get(lut_name, presets["cinematic"]) for b in range(out.shape[0]): img = out[b] # [H,W,3] # Luminance hesapla (mask için) lum = 0.299 * img[..., 0] + 0.587 * img[..., 1] + 0.114 * img[..., 2] shadow_m = torch.clamp(1.0 - lum * 2.5, 0, 1) highlight_m = torch.clamp((lum - 0.5) * 2.0, 0, 1) midtone_m = 1.0 - shadow_m - highlight_m midtone_m = torch.clamp(midtone_m, 0, 1) for c in range(3): channel = img[..., c] adj = (channel * (shadow_m * p["shadow"][c] + midtone_m * p["midtone"][c] + highlight_m * p["highlight"][c])) # Lift (vintage için raised blacks) lift = p.get("lift", 0.0) if lift > 0: adj = adj * (1 - lift) + lift # Contrast (0.5 etrafında ölçekle) contrast = p.get("contrast", 1.0) if contrast != 1.0: adj = (adj - 0.5) * contrast + 0.5 img[..., c] = adj # Saturation (HSV-benzeri lineer interpolation luma'ya) sat = p.get("saturation", 1.0) if sat != 1.0: lum = 0.299 * img[..., 0] + 0.587 * img[..., 1] + 0.114 * img[..., 2] for c in range(3): img[..., c] = lum + (img[..., c] - lum) * sat # Strength blend (original ile karıştır) if strength < 1.0: img = image_tensor[b] * (1 - strength) + img * strength out[b] = torch.clamp(img, 0, 1) return out, f"✓ LUT '{lut_name}' uygulandı (s={strength})" def _apply_vignette(image_tensor, intensity=0.3, falloff=0.6): """Köşelerde karartma — sinematik odak.""" try: import torch except ImportError: return image_tensor, "torch yok" if intensity <= 0: return image_tensor, "Vignette: kapalı" out = image_tensor.clone() _, h, w, _ = out.shape yy, xx = torch.meshgrid(torch.linspace(-1, 1, h), torch.linspace(-1, 1, w), indexing="ij") dist = torch.sqrt(xx ** 2 + yy ** 2) mask = 1.0 - torch.clamp((dist - falloff) / (1.4 - falloff), 0, 1) * intensity mask = mask.unsqueeze(-1) # [H,W,1] for b in range(out.shape[0]): out[b] = out[b] * mask return torch.clamp(out, 0, 1), f"✓ Vignette (i={intensity}, f={falloff})" def _apply_chromatic_aberration(image_tensor, intensity=2): """RGB kanal kaydırma — analog film hissi.""" try: import torch except ImportError: return image_tensor, "torch yok" if intensity <= 0: return image_tensor, "CA: kapalı" out = image_tensor.clone() sh = int(round(intensity)) if sh <= 0: return out, "CA: shift=0" for b in range(out.shape[0]): # Red kanalı sağa, Blue kanalı sola kaydır r = out[b, :, :, 0] bch = out[b, :, :, 2] r_shifted = torch.roll(r, shifts=sh, dims=1) b_shifted = torch.roll(bch, shifts=-sh, dims=1) out[b, :, :, 0] = r_shifted out[b, :, :, 2] = b_shifted return torch.clamp(out, 0, 1), f"✓ Chromatic aberration (shift={sh}px)" def _apply_letterbox(image_tensor, target_ratio=2.39): """Üst+alt siyah bar ekle — sinematik aspect.""" try: import torch except ImportError: return image_tensor, "torch yok" out = image_tensor.clone() _, h, w, _ = out.shape current_ratio = w / h if current_ratio >= target_ratio: return out, f"Letterbox gerek yok (ratio={current_ratio:.2f})" # Hedef yüksekliği hesapla new_h = int(round(w / target_ratio)) bar_h = (h - new_h) // 2 if bar_h <= 0: return out, "Letterbox: bar yok" for b in range(out.shape[0]): out[b, :bar_h, :, :] = 0.0 out[b, h - bar_h:, :, :] = 0.0 return out, f"✓ Letterbox {target_ratio:.2f}:1 (bar={bar_h}px)" class FikoYT_MVD_ProfessionalGrade: """Sprint 6: Profesyonel müzik klibi post-prodüksiyon paketi. CinematicPost'u (film_grain + sharpen + color_match) tamamlar: - 7 LUT preset (cinematic, vintage, vibrant, teal_orange, noir, warm, cool) - Vignette (intensity + falloff) - Chromatic aberration (analog film) - Letterbox (sinematik aspect, 2.39/2.35) Tüm efektler sırayla uygulanır: LUT → CA → Vignette → Letterbox. Pipeline'da CinematicPost'tan SONRA bağlanmalı (grain/sharpen → grade). """ CATEGORY = CATEGORY FUNCTION = "grade" RETURN_TYPES = ("IMAGE", "STRING") RETURN_NAMES = ("image", "report") @classmethod def INPUT_TYPES(cls): return { "required": { "image": ("IMAGE",), "lut_preset": (["none", "cinematic", "vintage", "vibrant", "teal_orange", "noir", "warm", "cool"], {"default": "cinematic"}), "lut_strength": ("FLOAT", {"default": 0.8, "min": 0.0, "max": 1.0, "step": 0.05}), "vignette_intensity": ("FLOAT", {"default": 0.25, "min": 0.0, "max": 1.0, "step": 0.05, "tooltip": "0=off, 0.2-0.4 hafif, 0.6+ ağır"}), "vignette_falloff": ("FLOAT", {"default": 0.6, "min": 0.0, "max": 1.0, "step": 0.05}), "chromatic_aberration": ("INT", {"default": 0, "min": 0, "max": 8, "step": 1, "tooltip": "RGB kanal kaydırma piksel — 1-2 hafif analog, 5+ heavy distortion"}), "letterbox_ratio": ("FLOAT", {"default": 0.0, "min": 0.0, "max": 3.0, "step": 0.01, "tooltip": "0=off, 2.35 klasik sinematik, 2.39 modern, 2.76 IMAX"}), } } def grade(self, image, lut_preset, lut_strength, vignette_intensity, vignette_falloff, chromatic_aberration, letterbox_ratio): notes = [] out = image if lut_preset != "none": out, msg = _apply_lut_3d(out, lut_preset, lut_strength) notes.append(msg) if chromatic_aberration > 0: out, msg = _apply_chromatic_aberration(out, chromatic_aberration) notes.append(msg) if vignette_intensity > 0: out, msg = _apply_vignette(out, vignette_intensity, vignette_falloff) notes.append(msg) if letterbox_ratio > 0: out, msg = _apply_letterbox(out, letterbox_ratio) notes.append(msg) return (out, "\n".join(notes) if notes else "ProfessionalGrade: tüm efektler kapalı") class FikoYT_MVD_AspectPreset: """Müzik klibi platform preset'leri — TikTok/YouTube/Insta için tek-tıkla boyutlandırma. Çıktı: width, height, label, ratio_str. Wizard veya EmptyLatentImage'a bağlanır. Strategy: contain (siyah padding) | cover (crop) | stretch. """ CATEGORY = CATEGORY FUNCTION = "resolve" RETURN_TYPES = ("INT", "INT", "STRING", "STRING") RETURN_NAMES = ("width", "height", "label", "ratio") @classmethod def INPUT_TYPES(cls): return { "required": { "preset": (list(MUSIC_VIDEO_ASPECT_PRESETS.keys()), {"default": "youtube"}), } } def resolve(self, preset): w, h, label, _platform = MUSIC_VIDEO_ASPECT_PRESETS.get( preset, MUSIC_VIDEO_ASPECT_PRESETS["youtube"]) from math import gcd g = gcd(w, h) ratio_str = f"{w // g}:{h // g}" return (w, h, label, ratio_str) class FikoYT_MVD_MasterRender: """FFmpeg ile tüm segment video'larını + audio'yu birleştir → master MP4. Input: segment video path listesi (JSON string veya project dizini) Output: master video path + report. Aspect ratio normalize (target_w/h ile pad veya crop), audio track ekle, chapter markers (segment başlangıçları), fade-in/out, watermark. """ CATEGORY = CATEGORY FUNCTION = "render" RETURN_TYPES = ("STRING", "STRING") RETURN_NAMES = ("output_path", "report") @classmethod def INPUT_TYPES(cls): return { "required": { "project_dir": ("STRING", {"default": "", "multiline": False, "placeholder": "ComfyUI/output/fikoyt_mvd/_"}), "audio_path": ("STRING", {"default": "", "multiline": False, "placeholder": "Master ses dosyası (mp3/wav)"}), "video_subdir": ("STRING", {"default": "videos", "multiline": False}), "output_filename": ("STRING", {"default": "master.mp4", "multiline": False}), "target_width": ("INT", {"default": 1920, "min": 256, "max": 7680, "step": 8}), "target_height": ("INT", {"default": 1080, "min": 256, "max": 4320, "step": 8}), "crf": ("INT", {"default": 18, "min": 0, "max": 51, "step": 1, "tooltip": "0=lossless, 18=görsel-lossless, 23=default, 28=küçük"}), "preset": (["ultrafast", "superfast", "veryfast", "faster", "fast", "medium", "slow", "slower", "veryslow"], {"default": "medium"}), "fade_in": ("FLOAT", {"default": 0.5, "min": 0.0, "max": 5.0, "step": 0.1}), "fade_out": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 5.0, "step": 0.1}), } } def render(self, project_dir, audio_path, video_subdir, output_filename, target_width, target_height, crf, preset, fade_in, fade_out): from ._legacy import _ffmpeg_bin import os import subprocess import tempfile ffmpeg = _ffmpeg_bin() if not ffmpeg: return ("", "❌ FFmpeg bulunamadı (PATH veya imageio_ffmpeg)") if not os.path.isdir(project_dir): return ("", f"❌ project_dir yok: {project_dir}") vdir = os.path.join(project_dir, video_subdir) if not os.path.isdir(vdir): return ("", f"❌ video dir yok: {vdir}") # Segment video'larını topla (sorted) segs = sorted([os.path.join(vdir, f) for f in os.listdir(vdir) if f.lower().endswith((".mp4", ".mov", ".mkv", ".webm"))]) if not segs: return ("", f"❌ {vdir} altında video yok") out_path = os.path.join(project_dir, output_filename) # FFmpeg concat list dosyası with tempfile.NamedTemporaryFile(mode="w", suffix=".txt", delete=False, encoding="utf-8") as f: for v in segs: # Escape single quotes for FFmpeg concat v_esc = v.replace("\\", "/").replace("'", r"'\''") f.write(f"file '{v_esc}'\n") list_file = f.name notes = [f"✓ {len(segs)} segment video bulundu"] try: # 1) Önce segment'leri concat et (yeniden encode etmeden) tmp_concat = os.path.join(project_dir, ".tmp_concat.mp4") concat_cmd = [ ffmpeg, "-y", "-f", "concat", "-safe", "0", "-i", list_file, "-c", "copy", tmp_concat ] r = subprocess.run(concat_cmd, capture_output=True, text=True) if r.returncode != 0: # Re-encode fallback (segment'ler farklı codec'lerde olabilir) notes.append("⚠ copy concat fail, re-encode'e geçildi") concat_cmd = [ ffmpeg, "-y", "-f", "concat", "-safe", "0", "-i", list_file, "-vf", f"scale={target_width}:{target_height}:force_original_aspect_ratio=decrease," f"pad={target_width}:{target_height}:(ow-iw)/2:(oh-ih)/2:black", "-c:v", "libx264", "-preset", preset, "-crf", str(crf), "-pix_fmt", "yuv420p", tmp_concat ] r = subprocess.run(concat_cmd, capture_output=True, text=True) if r.returncode != 0: return ("", f"❌ concat fail: {r.stderr[-500:]}") # 2) Audio + final pass (scale + pad + fade) video_filter = ( f"scale={target_width}:{target_height}:force_original_aspect_ratio=decrease," f"pad={target_width}:{target_height}:(ow-iw)/2:(oh-ih)/2:black" ) if fade_in > 0: video_filter += f",fade=t=in:st=0:d={fade_in}" if fade_out > 0: # Toplam süreyi öğrenmek için ffprobe try: probe = subprocess.run( [ffmpeg.replace("ffmpeg", "ffprobe"), "-v", "error", "-show_entries", "format=duration", "-of", "default=noprint_wrappers=1:nokey=1", tmp_concat], capture_output=True, text=True) duration = float(probe.stdout.strip()) video_filter += f",fade=t=out:st={duration - fade_out}:d={fade_out}" except Exception: pass final_cmd = [ffmpeg, "-y", "-i", tmp_concat] audio_args = [] if audio_path and os.path.isfile(audio_path): final_cmd.extend(["-i", audio_path]) audio_args = ["-c:a", "aac", "-b:a", "192k", "-shortest", "-map", "0:v:0", "-map", "1:a:0"] notes.append(f"✓ Audio eklendi: {os.path.basename(audio_path)}") else: audio_args = ["-an"] notes.append("⚠ Audio yok (audio_path boş veya yok)") final_cmd.extend([ "-vf", video_filter, "-c:v", "libx264", "-preset", preset, "-crf", str(crf), "-pix_fmt", "yuv420p", ] + audio_args + [out_path]) r = subprocess.run(final_cmd, capture_output=True, text=True) if r.returncode != 0: return ("", f"❌ final render fail: {r.stderr[-500:]}") # Cleanup try: os.remove(tmp_concat) os.remove(list_file) except Exception: pass file_mb = os.path.getsize(out_path) / 1_048_576 notes.append(f"✓ Master render: {output_filename} ({file_mb:.1f} MB)") return (out_path, "\n".join(notes)) except Exception as e: return ("", f"❌ MasterRender hata: {e}") # ============================================================================= # NODE_CLASS_MAPPINGS — Sprint 2/3/5/6 yeni node'ları registry'ye ekle # ============================================================================= # ============================================================================= # Sprint 16 (Flux Kontext): Multi-image karakter sürekliliği # ============================================================================= class FikoYT_MVD_FluxKontextChain: """🔗 Flux Kontext: multi-image referans + text edit ile karakter sürekliliği. BFL Flux.1 Kontext Dev (12B DiT). Önceki segment görselini "kontext" olarak alır, yeni prompt ile aynı karakter/sahne/stil tutarlılığını korur. Use case: Sahne 1 render edildi → Sahne 2'yi render ederken Sahne 1 görselini referans olarak ver → karakter, kıyafet, stil tutarlı kalır. Node feature-detect: 'ReferenceLatent' veya 'FluxKontextImageRescale' varsa Kontext zinciri kurulur. Yoksa graceful uyarı. """ CATEGORY = CATEGORY FUNCTION = "chain" RETURN_TYPES = ("IMAGE", "STRING") RETURN_NAMES = ("image", "status") @classmethod def INPUT_TYPES(cls): return { "required": { "kontext_image": ("IMAGE", { "tooltip": "Önceki segment görseli (karakter sürekliliği için referans)" }), "model": ("MODEL",), "clip": ("CLIP",), "vae": ("VAE",), "edit_prompt": ("STRING", { "default": "the same character, now in a different scene, " "maintaining the same identity, outfit, and style", "multiline": True, "tooltip": "Edit komutu — karakter aynı kalsın, sahne değişsin" }), "negative_prompt": ("STRING", { "default": "blurry, deformed, different person, watermark", "multiline": True, }), "width": ("INT", {"default": 1024, "min": 256, "max": 4096, "step": 64}), "height": ("INT", {"default": 1024, "min": 256, "max": 4096, "step": 64}), "steps": ("INT", {"default": 22, "min": 4, "max": 50, "step": 1}), "cfg": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 10.0, "step": 0.5}), "flux_guidance": ("FLOAT", {"default": 2.5, "min": 0.0, "max": 10.0, "step": 0.1, "tooltip": "Kontext default 2.5 (Flux Dev 3.5'den düşük)"}), "seed": ("INT", {"default": 0, "min": 0, "max": 0xFFFFFFFF}), }, "optional": { "second_reference": ("IMAGE", { "tooltip": "İsteğe bağlı 2. referans (örn: hedef poz/kıyafet)" }), } } def chain(self, kontext_image, model, clip, vae, edit_prompt, negative_prompt, width, height, steps, cfg, flux_guidance, seed, second_reference=None): try: from nodes import (NODE_CLASS_MAPPINGS as _NCM, CLIPTextEncode, EmptyLatentImage, KSampler, VAEDecode) import random except ImportError as e: return (_empty_image(), f"❌ import: {e}") # Kontext-specific node'ları feature-detect ref_latent_cls = _NCM.get("ReferenceLatent") # ComfyUI Kontext native kontext_rescale_cls = _NCM.get("FluxKontextImageRescale") if not ref_latent_cls and not kontext_rescale_cls: return (_empty_image(), "❌ Flux Kontext node'ları yok. ComfyUI 0.3.30+ veya " "ComfyUI-FluxKontextCreator yükle. Workflow template'i: " "workflow/flux_kontext_segment.json") if seed == 0: seed = random.randint(1, 0xFFFFFFFF) try: # Text encode cte = CLIPTextEncode() pos = cte.encode(clip, edit_prompt)[0] neg = cte.encode(clip, negative_prompt)[0] # Flux guidance fg_cls = _NCM.get("FluxGuidance") if fg_cls: pos = fg_cls().append(pos, flux_guidance)[0] # Kontext image rescale (önerilen) ctx_img = kontext_image if kontext_rescale_cls: try: ctx_img = kontext_rescale_cls().rescale(kontext_image)[0] except Exception: pass # VAE encode kontext image → latent vae_encode_cls = _NCM.get("VAEEncode") if not vae_encode_cls: return (_empty_image(), "❌ VAEEncode node yok") ctx_latent = vae_encode_cls().encode(vae, ctx_img)[0] # ReferenceLatent ile positive cond'a kontext bağla if ref_latent_cls: pos = ref_latent_cls().append(pos, ctx_latent)[0] # 2. referans (opsiyonel) — ikinci kontext if second_reference is not None and ref_latent_cls: ctx2 = second_reference if kontext_rescale_cls: try: ctx2 = kontext_rescale_cls().rescale(second_reference)[0] except Exception: pass ctx2_latent = vae_encode_cls().encode(vae, ctx2)[0] pos = ref_latent_cls().append(pos, ctx2_latent)[0] # Empty latent for new generation eli = EmptyLatentImage() latent = eli.generate(width, height, 1)[0] # Sample ks = KSampler() samples = ks.sample(model, seed, steps, cfg, "euler", "beta", pos, neg, latent, 1.0)[0] vd = VAEDecode() out = vd.decode(vae, samples)[0] return (out, f"✓ Flux Kontext zinciri uygulandı " f"({width}×{height}, steps={steps}, g={flux_guidance})") except Exception as e: import traceback; traceback.print_exc() return (_empty_image(), f"❌ Kontext chain hata: {e}") NODE_CLASS_MAPPINGS_VISUAL_EXTRA = { "FikoYT_MVD_PoseExtractor": FikoYT_MVD_PoseExtractor, "FikoYT_MVD_ControlNetUnion": FikoYT_MVD_ControlNetUnion, "FikoYT_MVD_RegionalMultiChar": FikoYT_MVD_RegionalMultiChar, "FikoYT_MVD_LightningPreview": FikoYT_MVD_LightningPreview, "FikoYT_MVD_ProfessionalGrade": FikoYT_MVD_ProfessionalGrade, "FikoYT_MVD_AspectPreset": FikoYT_MVD_AspectPreset, "FikoYT_MVD_MasterRender": FikoYT_MVD_MasterRender, "FikoYT_MVD_FluxKontextChain": FikoYT_MVD_FluxKontextChain, } NODE_DISPLAY_NAME_MAPPINGS_VISUAL_EXTRA = { "FikoYT_MVD_PoseExtractor": "🦴 MVD · Pose Extractor (DWPose)", "FikoYT_MVD_ControlNetUnion": "🎛 MVD · ControlNet Union", "FikoYT_MVD_RegionalMultiChar": "🧑‍🤝‍🧑 MVD · Regional Multi-Char", "FikoYT_MVD_LightningPreview": "⚡ MVD · Lightning Preview (4-step)", "FikoYT_MVD_ProfessionalGrade": "🎨 MVD · Professional Grade (LUT/Vignette/CA/Letterbox)", "FikoYT_MVD_AspectPreset": "📐 MVD · Aspect Preset (TikTok/YT/Insta)", "FikoYT_MVD_MasterRender": "🎬 MVD · Master Render (FFmpeg + audio)", "FikoYT_MVD_FluxKontextChain": "🔗 MVD · Flux Kontext Chain (segment sürekliliği)", } __all__ = [ "FikoYT_MVD_SegmentImageGen", "FikoYT_MVD_HiResFix", "FikoYT_MVD_AlbumCoverGen", "FikoYT_MVD_ImageAspectBatch", "FikoYT_MVD_PoseExtractor", "FikoYT_MVD_ControlNetUnion", "FikoYT_MVD_RegionalMultiChar", "FikoYT_MVD_LightningPreview", "FikoYT_MVD_ProfessionalGrade", "FikoYT_MVD_AspectPreset", "FikoYT_MVD_MasterRender", "MUSIC_VIDEO_ASPECT_PRESETS", "_empty_image", "_get_model_search_paths", "_try_apply_flux_guidance", "_try_apply_pulid", "_load_input_image", "_try_apply_ipadapter", "_try_run_preprocessor", "_try_apply_controlnet", "_try_apply_controlnet_union", "_classify_ref_mode", "_apply_references_chain", "_get_missing_flux_models", "_download_one_sync", "_comfy_class_available", "_find_controlnet_file", "_visual_ref_capabilities", "_detect_pipeline_nodes", "_next_free_id", "_inject_visual_references", "NODE_CLASS_MAPPINGS_VISUAL_EXTRA", "NODE_DISPLAY_NAME_MAPPINGS_VISUAL_EXTRA", ]