#!/usr/bin/env python3 """ Centralized configuration for video-replicator scripts. All constants, paths, and presets are defined here instead of being scattered across individual scripts. Import from this module: from config import ( VEO_CLI_PATH, PROJECT_BASE, LANDSCAPE_WIDTH, LANDSCAPE_HEIGHT, PORTRAIT_WIDTH, PORTRAIT_HEIGHT, ) """ import os from pathlib import Path # ============================================================================ # Path Resolution # ============================================================================ # scripts/ directory SCRIPTS_DIR = Path(__file__).resolve().parent # videoclaw project root (2 levels up from scripts/video/) _PROJECT_ROOT = SCRIPTS_DIR.parent.parent # Default paths (overridable via environment variables) VEO_CLI_PATH: str = os.environ.get( "VEO_CLI_PATH", str(_PROJECT_ROOT / "veo-cli") ) PROJECT_BASE: str = os.environ.get( "VIDEO_REPLICATOR_PROJECTS", str(Path.home() / "videoclaw" / "projects") ) # ============================================================================ # Image Dimensions # ============================================================================ # Standard portrait dimensions (9:16 ratio) PORTRAIT_WIDTH: int = 504 PORTRAIT_HEIGHT: int = 896 PORTRAIT_RATIO: float = 9 / 16 # 0.5625 # Standard landscape dimensions (16:9 ratio) — matches Veo output LANDSCAPE_WIDTH: int = 1280 LANDSCAPE_HEIGHT: int = 720 LANDSCAPE_RATIO: float = 16 / 9 # 1.778 # ============================================================================ # Retry Settings # ============================================================================ MAX_RETRIES: int = 2 RETRY_DELAY_SECONDS: int = 30 # Diagnostic prompt for image rejection detection (v2.39) # Used to test whether an image is being rejected by Google's content filter. # This prompt is intentionally neutral — no subject matter, no action, just # camera movement and lighting. If generation fails with this prompt too, # the IMAGE itself is the problem (not the prompt text). IMAGE_DIAGNOSTIC_PROMPT: str = "Slow camera push in, ambient light, cinematic" # ============================================================================ # File Size Thresholds # ============================================================================ # Minimum valid video file size (100KB) — smaller files are likely corrupt MIN_VIDEO_SIZE_BYTES: int = 100_000 # Minimum valid image file size (1KB) MIN_IMAGE_SIZE_BYTES: int = 1_000 # ============================================================================ # Audio Volume Presets # ============================================================================ AUDIO_PRESETS: dict = { "default": { "music_volume": 0.6, "video_volume": 0.3, "music_fade_out": 3.0, }, "presenter": { "music_volume": 0.25, "video_volume": 0.85, "music_fade_out": 3.0, }, "narrated": { "music_volume": 0.15, "video_volume": 0.85, "music_fade_out": 3.0, }, } # ============================================================================ # Quality Presets # ============================================================================ QUALITY_PRESETS: dict = { "draft": { "quality": "fast", "variations": 1, }, "final": { "quality": "quality", "variations": 2, }, } # Credits per video by quality tier CREDITS_PER_VIDEO: dict = { "quality": 100, "fast": 10, "free": 0, "veo2": 100, } # Approximate cost per credit (USD) COST_PER_CREDIT: float = 0.005 # ============================================================================ # Lip-Sync Prompt Pattern # ============================================================================ # Template for Veo lip-sync dialogue scenes. # Proven pattern: Veo 3 generates lip-synced audio when dialogue is in the prompt. # {character} = character description, {dialogue} = spoken text, # {action} = physical actions, {environment} = scene setting LIP_SYNC_PROMPT_PATTERN: str = ( "The {character} speaks directly to camera saying: \"{dialogue}\". " "{action}. {environment}." ) # ============================================================================ # FFmpeg Settings # ============================================================================ FFMPEG_TIMEOUT: int = 300 # 5 minutes default timeout FFPROBE_TIMEOUT: int = 10 CONCAT_FILTER_THRESHOLD: int = 8 # Use concat filter (not demuxer) for 8+ segments to prevent A/V drift # FPS Normalization (v2.45) — explicit constants for Seedance/Veo output frame rates SEEDANCE_OUTPUT_FPS: int = 60 # Seedance outputs 60fps TARGET_ASSEMBLY_FPS: int = 24 # Pipeline normalizes everything to 24fps # ============================================================================ # UGC Campaign Settings # ============================================================================ # Default number of belief scripts per campaign UGC_DEFAULT_SCRIPTS: int = 3 # Scene types in UGC belief-driven scripts UGC_SCENE_TYPES: list = [ "hook", "problem", "mechanism", "proof", "offer", "cta", ] # Word limits per segment duration (seconds → max words) UGC_WORD_LIMITS: dict = { 4: 12, 6: 18, 8: 25, 12: 35, } # Subtitle defaults UGC_SUBTITLE_FONT_SIZE: int = 48 UGC_SUBTITLE_WORDS_PER_LINE: int = 6 # UGC audio preset (voice-forward with subtle music) AUDIO_PRESETS["ugc"] = { "music_volume": 0.10, "video_volume": 0.90, "music_fade_out": 2.0, } # ============================================================================ # Seedance 2.0 Backend Settings # ============================================================================ SEEDANCE_BASE_URL: str = "https://api.xskill.ai" SEEDANCE_CREATE_URL: str = f"{SEEDANCE_BASE_URL}/api/v3/tasks/create" SEEDANCE_QUERY_URL: str = f"{SEEDANCE_BASE_URL}/api/v3/tasks/query" SEEDANCE_CANCEL_URL: str = f"{SEEDANCE_BASE_URL}/api/v3/tasks/cancel" SEEDANCE_BALANCE_URL: str = f"{SEEDANCE_BASE_URL}/api/v3/balance" # ark/seedance-2.0 (default) vs legacy st-ai/super-seed2 # Set SEEDANCE_USE_ARK_API=false to revert to the legacy model. SEEDANCE_USE_ARK_API: bool = os.environ.get("SEEDANCE_USE_ARK_API", "true").lower() != "false" SEEDANCE_MODEL_ID_ARK: str = "ark/seedance-2.0" SEEDANCE_MODEL_ID_LEGACY: str = "st-ai/super-seed2" SEEDANCE_MODEL_ID: str = SEEDANCE_MODEL_ID_ARK if SEEDANCE_USE_ARK_API else SEEDANCE_MODEL_ID_LEGACY # ark/seedance-2.0 specific defaults SEEDANCE_DEFAULT_RESOLUTION: str = "720p" SEEDANCE_GENERATE_AUDIO: bool = True SEEDANCE_WATERMARK: bool = False SEEDANCE_QUALITY_MAP: dict = { "fast": "seedance_2.0_fast", "quality": "seedance_2.0", } SEEDANCE_RATIO_MAP: dict = { "landscape": "16:9", "portrait": "9:16", "square": "1:1", } SEEDANCE_MIN_DURATION: int = 4 SEEDANCE_MAX_DURATION: int = 15 SEEDANCE_DEFAULT_DURATION: int = 15 SEEDANCE_POLL_INTERVAL: int = 5 # seconds between status polls SEEDANCE_MAX_POLL_TIME: int = 1200 # 20 minutes max wait per task CATBOX_UPLOAD_URL: str = "https://catbox.moe/user/api.php" UGUU_UPLOAD_URL: str = "https://uguu.se/upload" UGUU_MAX_FILE_SIZE: int = 100 * 1024 * 1024 # 100MB (uguu.se limit) # xskill.ai upload (uses same CDN as Seedance — guaranteed accessible) XSKILL_UPLOAD_URL: str = f"{SEEDANCE_BASE_URL}/api/v3/upload" XSKILL_MAX_FILE_SIZE: int = 10 * 1024 * 1024 # 10MB conservative limit # xskill.ai MCP upload (Streamable HTTP, works when /api/v3/upload returns 404) XSKILL_MCP_HTTP_URL: str = f"{SEEDANCE_BASE_URL}/api/v3/mcp-http" XSKILL_MCP_MESSAGE_URL: str = f"{SEEDANCE_BASE_URL}/api/v3/mcp/message" # legacy SSE-era path (fallback) XSKILL_MCP_TIMEOUT: int = 60 # upload timeout # Seedance quality fallback (model unavailability recovery) SEEDANCE_FALLBACK_QUALITY: dict = { "seedance_2.0": "seedance_2.0_fast", # quality → fast } SEEDANCE_MODEL_NOT_FOUND_PATTERNS: list = [ "模型菜单中未找到", # Chinese: "model not found in menu" "model not found", "model unavailable", ] # Seedance per-mode credit costs (flat per-5s, legacy — kept for backward compat) SEEDANCE_CREDITS_PER_MODE: dict = { "text-to-video": {"fast": 12, "quality": 19}, "frames-to-video": {"fast": 19, "quality": 19}, "audio-lipsync": {"fast": 19, "quality": 19}, "motion-transfer": {"fast": 38, "quality": 38}, "camera-ref": {"fast": 24, "quality": 24}, } # Seedance per-second credit rates (upstream docs, duration-aware pricing) # Modes without video input: fast=2.4/sec, quality=3.0/sec # Modes with video input: fast=4.8/sec, quality=6.0/sec SEEDANCE_CREDITS_PER_SECOND: dict = { "text-to-video": {"fast": 2.4, "quality": 3.0}, "frames-to-video": {"fast": 2.4, "quality": 3.0}, "audio-lipsync": {"fast": 2.4, "quality": 3.0}, "motion-transfer": {"fast": 4.8, "quality": 6.0}, "camera-ref": {"fast": 4.8, "quality": 6.0}, } # 4K Upscale Pre-Flight (v2.43) — upscale frame images via Go Bananas before Seedance UPSCALE_4K_DEFAULT_PROMPT: str = "upscale image, high detail, 4K, preserve all details, cinematic quality" # Style Consistency Classification (v2.43) — Gemini Vision style categories STYLE_CONSISTENCY_CATEGORIES: list = [ "photorealistic", "3d_animated", "anime", "sketch", "illustration", "mixed", ] # Seedance output resolution (xskill.ai platform limitation) SEEDANCE_OUTPUT_RESOLUTION: str = "640x360" SEEDANCE_UPSCALE_RESOLUTION: str = "1280x720" # Seedance duration validation SEEDANCE_DURATION_TOLERANCE: float = 0.8 # retry if actual < 80% of requested SEEDANCE_OVER_DURATION_TOLERANCE: float = 1.1 # auto-trim if actual > 110% of requested # amix filter safe default (prevents early audio termination) AMIX_DROPOUT_TRANSITION: int = 600 # Domains known to be inaccessible from Seedance's Chinese infrastructure. # URLs on these domains are automatically re-downloaded and re-uploaded to a # China-accessible CDN (imgbb or xskill) before being sent to the Seedance API. SEEDANCE_REHOST_DOMAINS: list = [ "catbox.moe", # catbox uploads (blocked from China) "uguu.se", # uguu temporary uploads (expire ~48h, not China-reliable) # Note: r2.dev (Cloudflare R2) removed — Go Bananas R2 URLs are globally accessible ] # Seedance Dead Task Detection (v2.45) — detect stalled tasks in poll loop SEEDANCE_DEAD_TASK_THRESHOLD: int = 1200 # seconds with no updated_at change → assume dead (I2V can take 10-20 min) # Seedance Smart Polling (v2.37) — initial delay then longer intervals SEEDANCE_INITIAL_POLL_DELAY: int = 30 # Wait 30s before first poll (task won't be done sooner) SEEDANCE_SMART_POLL_INTERVAL: int = 15 # Then poll every 15s (was 5s, saves ~10 wasted calls) # Seedance Extended Ratio Map (v2.37) — includes ultrawide, classic, cinemascope # API-validated: only these 6 raw values accepted: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 SEEDANCE_RATIO_MAP_EXTENDED: dict = { "landscape": "16:9", "portrait": "9:16", "square": "1:1", "ultrawide": "21:9", "classic": "4:3", "classic_portrait": "3:4", "cinemascope": "21:9", # API rejects 2.35:1 — maps to closest supported ultrawide "adaptive": "adaptive", } # Seedance Media File Limits (v2.37) SEEDANCE_MAX_IMAGES: int = 9 SEEDANCE_MAX_VIDEOS: int = 3 SEEDANCE_MAX_AUDIO: int = 3 SEEDANCE_MAX_MEDIA_TOTAL: int = 12 SEEDANCE_MAX_REF_DURATION: int = 15 # max total seconds for video/audio references SEEDANCE_MAX_IMAGE_SIZE: int = 30 * 1024 * 1024 # 30MB per image SEEDANCE_IMAGE_MIN_RES: int = 300 # minimum image dimension (px) SEEDANCE_IMAGE_MAX_RES: int = 6000 # maximum image dimension (px) # Seedance Service Tiers (v2.37) — flex tier gives 50% cost savings SEEDANCE_SERVICE_TIERS: dict = { "default": {"cost_multiplier": 1.0}, "flex": {"cost_multiplier": 0.5}, } # Seedance content violation auto-retry (v2.38) SEEDANCE_MAX_CONTENT_RETRIES: int = 2 # Max sanitization retries per scene # Seedance auto-fallback backend (v2.45) — switch to Veo on permanent Seedance failure SEEDANCE_FALLBACK_BACKEND: str = "useapi" # Backend to fall back to SEEDANCE_FALLBACK_ENABLED: bool = False # Disabled by default; enable via --fallback-backend # Seedance pre-submission image analysis (v2.45) — Gemini-based risk detection SEEDANCE_PRE_ANALYZE_IMAGES: bool = False # Disabled by default; enable via --pre-analyze-images SEEDANCE_RISK_THRESHOLD: str = "high" # Block images at this risk level or above # Seedance Omni Reference Mode (v2.39) SEEDANCE_OMNI_FUNCTION_MODE: str = "omni_reference" SEEDANCE_OMNI_MAX_IMAGES: int = 4 # Max image_file_N slots (1-4) SEEDANCE_OMNI_MAX_DURATION: int = 15 # Single omni_reference max seconds SEEDANCE_OMNI_SEGMENT_DURATION: int = 15 # Max seconds per segment (for chaining) SEEDANCE_OMNI_MAX_SEGMENTS: int = 4 # Max chained segments (60s total) SEEDANCE_OMNI_IMAGE_ROLES: list = ["product", "environment", "character", "logo"] SEEDANCE_OMNI_CROSSFADE_DURATION: float = 0.5 # Crossfade between chained segments SEEDANCE_OMNI_CREDITS_PER_SEGMENT: int = 150 # Actual cost per omni_reference segment # Storyboard Panels JSON (v2.43) — maps scene numbers to panel images for Seedance reference STORYBOARD_PANELS_FILE: str = "analysis/storyboard_panels.json" # Camera Variety Block (v2.43) — append to Seedance prompts for dynamic camera angles CAMERA_VARIETY_BLOCK: str = ( "a lot of camera shots: Extreme wide shot, wide shot, medium shot, medium close-up, " "close-up, extreme close-up, over-the-shoulder, point-of-view, insert shot, two-shot, " "tracking shot, push-in, pull-back, crane shot, arc shot, high angle, low angle, " "Dutch angle, and top-down bird's-eye shot." ) # Seedance Prompt Engine (v2.35) SEEDANCE_NEGATIVE_PROMPTS: str = "No text, no subtitles, no watermarks, no logos, no abrupt cuts" SEEDANCE_TIME_SEGMENT_THRESHOLD: int = 8 # Add time segments for clips >= this duration # Seedance Prompt Library DB (v2.36) SEEDANCE_PROMPT_DB_PATH: str = str(SCRIPTS_DIR / "data" / "seedance_prompts.db") SEEDANCE_PROMPT_DATA_DIR: str = str(SCRIPTS_DIR / "data") # Seedance Batch Settings SEEDANCE_BATCH_SUBMIT_DELAY: float = 0.5 # seconds between submissions SEEDANCE_BATCH_MAX_WORKERS: int = 20 # max concurrent pollers SEEDANCE_BATCH_POLL_TIMEOUT: int = 1800 # 30 min default per task SEEDANCE_BATCH_CHECKPOINT_INTERVAL: int = 1 # save after every N completions # ============================================================================ # Seedance Webhook Settings (v2.46) # ============================================================================ SEEDANCE_WEBHOOK_ENABLED: bool = os.environ.get("SEEDANCE_WEBHOOK_ENABLED", "true").lower() != "false" SEEDANCE_WEBHOOK_PORT: int = int(os.environ.get("SEEDANCE_WEBHOOK_PORT", "0")) # 0 = auto SEEDANCE_WEBHOOK_USE_TUNNEL: bool = os.environ.get("SEEDANCE_WEBHOOK_USE_TUNNEL", "true").lower() != "false" SEEDANCE_WEBHOOK_PATH: str = "/webhook/seedance" # xskill.ai Asset Library (character reference images → Asset URIs bypass content filter) SEEDANCE_ASSET_GROUPS_URL: str = f"{SEEDANCE_BASE_URL}/api/v3/assets/groups" SEEDANCE_ASSETS_URL: str = f"{SEEDANCE_BASE_URL}/api/v3/assets" SEEDANCE_ASSET_GROUP_NAME: str = "videoclaw-characters" SEEDANCE_ASSET_CACHE_FILE: str = str(Path.home() / ".videoclaw" / "asset_cache.json") # ============================================================================ # useapi.net Backend Settings (Extend Chain, v2.38) # ============================================================================ USEAPI_BASE_URL: str = "https://api.useapi.net/v1" USEAPI_VIDEOS_URL: str = f"{USEAPI_BASE_URL}/google-flow/videos" USEAPI_EXTEND_URL: str = f"{USEAPI_BASE_URL}/google-flow/videos/extend" USEAPI_CONCAT_URL: str = f"{USEAPI_BASE_URL}/google-flow/videos/concatenate" USEAPI_ASSETS_URL: str = f"{USEAPI_BASE_URL}/google-flow/assets" # append /{email} USEAPI_JOBS_URL: str = f"{USEAPI_BASE_URL}/google-flow/jobs" # append /{jobId} USEAPI_MODEL_MAP: dict = { "fast": "veo-3.1-fast", "quality": "veo-3.1-quality", "lite": "veo-3.1-lite", "free": "veo-3.1-lite-low-priority", # veo-3.1-fast-relaxed was removed by the API "relaxed": "veo-3.1-lite-low-priority", "omni-flash": "omni-flash", } USEAPI_POLL_INTERVAL: int = 3 # seconds between job status polls USEAPI_MAX_POLL_TIME: int = 600 # 10 minutes max wait USEAPI_REQUEST_TIMEOUT: int = 30 # HTTP request timeout (non-generation) USEAPI_GENERATION_TIMEOUT: int = 180 # 3 minutes for video generation (sync response) USEAPI_DOWNLOAD_TIMEOUT: int = 120 # video download timeout USEAPI_MAX_IMAGE_SIZE: int = 20 * 1024 * 1024 # 20MB upload limit # Google Flow v1 video duration (seconds): 4/6 Ultra-only on Veo, 10 omni-flash only. USEAPI_DEFAULT_DURATION: int = 8 # The 30 Google Flow voice-narration presets (referenceAudio_1..5). USEAPI_VOICE_PRESETS: list = [ "Achird", "Achernar", "Algieba", "Algenib", "Alnilam", "Aoede", "Autonoe", "Callirrhoe", "Charon", "Despina", "Enceladus", "Erinome", "Fenrir", "Gacrux", "Iapetus", "Kore", "Laomedeia", "Leda", "Orus", "Puck", "Pulcherrima", "Rasalgethi", "Sadachbia", "Sadaltager", "Schedar", "Sulafat", "Umbriel", "Vindemiatrix", "Zephyr", "Zubenelgenubi", ] # Extend Chain defaults EXTEND_CHAIN_OVERLAP_TRIM: float = 1.0 # seconds to trim from extension starts EXTEND_CHAIN_SEGMENT_DURATION: float = 8.0 # approximate duration per segment # ============================================================================ # Story Engine & Film Pipeline Settings (v2.37) # ============================================================================ GEMINI_FLASH_MODEL: str = os.environ.get("GEMINI_FLASH_MODEL", "gemini-3-flash-preview") GEMINI_PRO_MODEL: str = os.environ.get("GEMINI_PRO_MODEL", "gemini-2.0-pro-exp") STORY_ENGINE_DURATIONS: list = [4, 6, 8] STORY_ENGINE_DEFAULT_SHOTS: int = 6 STORY_ENGINE_MAX_SHOTS: int = 20 # Film Pipeline phases FILM_PHASES: list = [ "concept", "analysis", "screenplay", "characters", "breakdown", "images", "videos", "audio", "stitch", "complete", ] FILM_SUPPORTED_BACKENDS: list = ["seedance", "direct", "useapi"] # ============================================================================ # Music Generation Settings (Kie.ai / Suno) # ============================================================================ KIE_API_BASE: str = "https://api.kie.ai" MUSIC_MAX_RETRIES: int = 2 # Default retry attempts on failure MUSIC_RETRY_BACKOFF: list = [30, 60] # Seconds between retries (exponential) MUSIC_POLL_INTERVAL: int = 5 # Seconds between status polls MUSIC_POLL_MAX_ATTEMPTS: int = 60 # Max poll attempts (5 min at 5s interval) MUSIC_POLL_TIMEOUT_MULTIPLIER: int = 2 # Multiply poll attempts on retry # Retryable failure patterns (case-insensitive substring match) MUSIC_RETRYABLE_ERRORS: list = [ "GENERATE_AUDIO_FAILED", "timed out", "timeout", "Generation timed out", ] # ============================================================================ # Narration Conductor # ============================================================================ # =================================================================== # Title Card # =================================================================== # Default hold duration (seconds) TITLE_CARD_DURATION: int = 4 # Target dimensions (must match video pipeline) TITLE_CARD_WIDTH: int = LANDSCAPE_WIDTH # 1280 TITLE_CARD_HEIGHT: int = LANDSCAPE_HEIGHT # 720 # FPS (must match video pipeline) TITLE_CARD_FPS: int = 24 # Dark band at bottom (semi-transparent overlay for text legibility) TITLE_CARD_BAND_TOP: int = 560 TITLE_CARD_BAND_OPACITY: int = 160 # 0-255 alpha # Title text settings TITLE_CARD_TITLE_FONT_SIZE: int = 58 TITLE_CARD_TITLE_COLOR: str = "white" TITLE_CARD_TITLE_OUTLINE_WIDTH: int = 2 # Subtitle text settings TITLE_CARD_SUBTITLE_FONT_SIZE: int = 28 TITLE_CARD_SUBTITLE_COLOR: str = "#C8DCFF" # Light blue # Font (PIL will fall back to default if not found) TITLE_CARD_FONT: str = "Arial Bold" TITLE_CARD_FONT_FALLBACK: str = "Helvetica" # =================================================================== # Narration Conductor # =================================================================== # Time to wait before narration starts in each scene (breathing room) CONDUCTOR_DELAY_S: float = 0.5 # Minimum silence between scenes when stitched CONDUCTOR_GAP_S: float = 0.3 # Fade-in duration on each TTS segment CONDUCTOR_FADE_IN_S: float = 0.15 # Fade-out duration on each TTS segment CONDUCTOR_FADE_OUT_S: float = 0.25 # Maximum speed-up factor when TTS exceeds available window CONDUCTOR_MAX_SPEED: float = 1.15 # Skip narration for scene if available window is shorter than this CONDUCTOR_MIN_WINDOW_S: float = 1.0 # ============================================================================ # Go Bananas Style Transfer # ============================================================================ # Prompt template for style transfer via Go Bananas reference groups. # Use with generate_image(image_to_edit_id=X, reference_group_id=Y, # reference_mode="style"). Replace {group_name} with the actual group name. STYLE_TRANSFER_PROMPT_TEMPLATE: str = ( "Using the visual style from the reference images " '(reference group(s) "{group_name}"), style transfer. ' "Match the artistic style, color palette, lighting, mood, " "and visual aesthetic from the references." ) # Character Reference Sheet presets for Go Bananas GO_BANANAS_CHAR_REF_SHEET_PRESET: int = 49 # 2x4 grid, 8 views GO_BANANAS_CINEMATIC_REF_PRESET: int = 55 # Cinematic, 7 views # Narration tone options for TTS script generation NARRATION_TONES: dict = { "conversational": "Write as if talking to a friend over coffee. First person, casual, honest. No jargon. Use contractions. Short sentences.", "corporate": "Professional business presentation tone. Clear, structured, data-driven. Use industry terminology appropriately.", "storytelling": "Narrative voice. Set scenes, build tension, reveal insights. Use 'I remember when...' and 'Here's what nobody tells you...' patterns.", "casual": "Very informal. Use colloquialisms, humor, rhetorical questions. Like a podcast conversation.", } DEFAULT_NARRATION_TONE: str = "conversational" # ============================================================================ # Genre Presets (cherry-picked from short-movie-maker) # ============================================================================ GENRE_PRESETS: dict = { "drama": { "pacing": "slow to medium", "avg_shot_duration_s": 6, "scenes_per_minute": 2.5, "music_bpm": 80, "music_prompt_keywords": "orchestral underscore, solo piano, minimal", "color_grading_ffmpeg": "eq=contrast=1.05:brightness=0.01:saturation=0.85", "transition": "dissolve", "transition_duration_s": 0.8, "lighting_keywords": "natural, motivated, low-key for emotional scenes", "camera_keywords": "steady, deliberate movements, long takes", }, "thriller": { "pacing": "builds from slow to fast", "avg_shot_duration_s": 4, "scenes_per_minute": 4, "music_bpm": 115, "music_prompt_keywords": "tense drones, staccato strings, building rhythm", "color_grading_ffmpeg": "eq=contrast=1.15:brightness=-0.02:saturation=0.75,colorbalance=rs=-0.1:bs=0.15", "transition": "fade", "transition_duration_s": 0.3, "lighting_keywords": "low-key, harsh shadows, motivated sources", "camera_keywords": "handheld for tension, steady for control", }, "comedy": { "pacing": "quick, rhythmic", "avg_shot_duration_s": 4, "scenes_per_minute": 3.5, "music_bpm": 135, "music_prompt_keywords": "playful, upbeat, quirky instruments", "color_grading_ffmpeg": "eq=contrast=1.05:brightness=0.03:saturation=1.15", "transition": "wipeleft", "transition_duration_s": 0.3, "lighting_keywords": "high-key, even, bright", "camera_keywords": "static or simple pans, quick zooms for punchlines", }, "sci-fi": { "pacing": "deliberate, with punctuated action", "avg_shot_duration_s": 5, "scenes_per_minute": 3, "music_bpm": 100, "music_prompt_keywords": "electronic, synth pads, atmospheric", "color_grading_ffmpeg": "eq=contrast=1.1:saturation=0.9,colorbalance=rs=-0.1:gs=-0.05:bs=0.2", "transition": "dissolve", "transition_duration_s": 0.5, "lighting_keywords": "practical neon, volumetric, rim lighting", "camera_keywords": "smooth tracking, crane shots, steady", }, "horror": { "pacing": "slow builds with sudden bursts", "avg_shot_duration_s": 6, "scenes_per_minute": 2.5, "music_bpm": 70, "music_prompt_keywords": "dissonant drones, creeping tension, sudden stingers", "color_grading_ffmpeg": "eq=contrast=1.2:brightness=-0.05:saturation=0.6,colorbalance=gs=0.1", "transition": "fade", "transition_duration_s": 1.0, "lighting_keywords": "extreme low-key, single source, deep shadows", "camera_keywords": "slow tracking, occasional POV, dutch angles", }, "romance": { "pacing": "gentle, flowing", "avg_shot_duration_s": 6, "scenes_per_minute": 2.5, "music_bpm": 80, "music_prompt_keywords": "gentle piano, acoustic guitar, soft strings", "color_grading_ffmpeg": "eq=contrast=0.95:brightness=0.03:saturation=1.05,colorbalance=rs=0.1:gs=0.05", "transition": "dissolve", "transition_duration_s": 1.0, "lighting_keywords": "golden hour, soft, diffused, backlit", "camera_keywords": "smooth dollies, slow pans, shallow DOF", }, "action": { "pacing": "fast, relentless", "avg_shot_duration_s": 3, "scenes_per_minute": 5, "music_bpm": 145, "music_prompt_keywords": "driving percussion, heavy bass, adrenaline", "color_grading_ffmpeg": "eq=contrast=1.2:brightness=0.0:saturation=1.1,colorbalance=rs=0.15:bs=-0.1", "transition": "fade", "transition_duration_s": 0.2, "lighting_keywords": "dynamic, practical explosions, rim lighting", "camera_keywords": "handheld, fast tracking, dynamic angles", }, } # Named color grade presets (usable independently of genre) COLOR_GRADE_PRESETS: dict = { "nolan": "eq=contrast=1.15:brightness=-0.02:saturation=0.8,colorbalance=rs=-0.05:bs=0.15", "warm-cinematic": "eq=contrast=1.05:brightness=0.02:saturation=1.05,colorbalance=rs=0.1:gs=0.05", "noir": "eq=contrast=1.25:brightness=-0.05:saturation=0.4", "pastel": "eq=contrast=0.9:brightness=0.05:saturation=0.75", } # ============================================================================ # Reference Scenarios (elementsix integration) # ============================================================================ # Maps scenario IDs to their Seedance 2.0 payload configuration. # Source: elementsix-skills research (2026-02-27) REFERENCE_SCENARIOS: dict[str, dict] = { "first_frame": { "description": "Use image as opening frame", "mode": "frames-to-video", "prompt_template": "@image1 {prompt}", "media_type": "image", "zh_equiv": "@图片1 作为首帧", }, "character_ref": { "description": "Character reference for face/body consistency", "mode": "omni_reference", "prompt_template": "@image_file_{N} as character reference. {prompt}", "media_type": "image", "zh_equiv": "@图片1 角色参考", }, "camera_clone": { "description": "Clone camera movement from reference video", "mode": "frames-to-video", "prompt_template": "{prompt} Fully reference @video1's camera movements.", "media_type": "video", "zh_equiv": "@视频1 运镜参考", }, "video_extend": { "description": "Extend existing video by N seconds", "mode": "video-extend", "prompt_template": "Continue @video1 seamlessly. {prompt}", "media_type": "video", "zh_equiv": "延长视频", }, "character_replace": { "description": "Replace character in video with new character from image", "mode": "character-replace", "prompt_template": "Replace main character in @video_file_1 with @image_file_1. {prompt}", "media_type": "image|video", "zh_equiv": "角色替换", }, "audio_clone": { "description": "Use audio as BGM/rhythm reference", "mode": "omni_reference", "prompt_template": "Use @audio_file_1 as background music reference. {prompt}", "media_type": "audio", "zh_equiv": "声音参考", }, } # ============================================================================ # Camera Vocabulary (bilingual Seedance-safe phrasing) # ============================================================================ # Maps English camera terms → Chinese equivalent → Seedance-safe prompt phrase → emotional energy. # Use these phrases in scene prompts for reliable Seedance camera move interpretation. CAMERA_VOCABULARY: dict[str, dict[str, str]] = { "push_in": {"zh": "推", "seedance": "camera slowly pushes forward", "energy": "approach"}, "pull_back": {"zh": "拉", "seedance": "camera pulls back to reveal", "energy": "reveal"}, "pan": {"zh": "摇", "seedance": "camera pans horizontally", "energy": "survey"}, "tilt": {"zh": "移", "seedance": "camera tilts vertically", "energy": "discover"}, "tracking": {"zh": "跟", "seedance": "tracking shot following subject", "energy": "follow"}, "orbit": {"zh": "环绕", "seedance": "camera orbits around subject", "energy": "inspect"}, "crane": {"zh": "升降", "seedance": "crane shot rising/descending", "energy": "grandeur"}, "hitchcock": {"zh": "希区柯克变焦", "seedance": "dolly zoom, background warps while subject stays fixed", "energy": "tension"}, }