Kinocut

Load a saved shorts plan and return source-free proposals for review.

Render platform drafts for an approved candidate from a saved plan.

Append one human review decision to a saved shorts plan (source-free).

Expand one storyboard shot into prompt and negative_prompt strings. Args: project_path: Absolute path to a project directory containing style.md and storyboard.md. shot: Shot id or 1-based row number from storyboard.md.

Discover the bounded public sound operation set (local-first, JSON-safe).

Retain a two-pass master only after measuring final audio against its policy.

Assemble supplied WAVs into a new ZIP; omit both arguments for a labelled demo.

Validate a SoundPlan payload (or a built-in minimal plan when omitted).

Recognize supplied local audio and compare it with a reference script.

Measure hashed local audio against its policy; omit inputs for a labelled demo.

Retain local caption speech from a request, or run the synthetic plan demo.

Fail-closed cohesion gate for a still package + contact sheet. Checks luma spread and shadow green/cyan wash. Returns passed=false with named metric/frame failures when the package fails. Args: inputs: Absolute still paths in the package. output_dir: Absolute directory for receipt and contact sheet.

Grade stills in order correct→match→look; optional 3D LUT applied last. Signal mode marks the LUT as signal-alignment (not film cosplay) and records near-black/near-white preservation deltas. Args: inputs: Absolute still paths. output_dir: Absolute output directory. hero: Optional hero still for match stage. lut_path: Optional absolute .cube/.3dl path applied last. signal_mode: Treat LUT as signal alignment.

Match a package of stills to a hero plate with one shared WB/exposure gain. Does not overwrite sources. Writes matched stills + JSON receipt under output_dir. Per-frame auto-WB is disabled (shared gains only). Args: hero: Absolute path to the hero/establish still. inputs: Absolute paths to package stills. output_dir: Absolute directory for outputs and receipt.

Multi-still package workflow: edit beats → match → grade → cohesion gate. Args: establish: Absolute establish/hero still. beats: Absolute beat still paths. output_dir: Absolute package output directory. dry_run: When True, return planned graph only. apply_grade: Run ordered grade after match (default True). lut_path: Optional LUT path for grade look stage. signal_mode: Signal-alignment LUT mode.

Read a PUSHING CREATION storyboard table from storyboard.md or a project directory. Args: path: Absolute path to storyboard.md or to a project directory containing storyboard.md.

Read STYLE_ and NEG_ blocks from a style.md file or project directory. Args: path: Absolute path to style.md or to a project directory containing style.md.

Apply glitch transition between two video clips. Args: clip1_path: Absolute path to first video clip. clip2_path: Absolute path to second video clip. output_path: Absolute path for output video. duration: Transition duration in seconds (default 0.5). intensity: Glitch intensity 0-1 (default 0.3).

Apply morph transition between two video clips.

Apply pixelate transition between two video clips.

Evaluate exact-spec verdict and defect evidence without approving anything.

Add, replace, or mix an audio track into an existing video and render a new output file. Returns JSON {success, output_path}. Use to score or re-voice a video; source video and audio are read only. Pass video_path as the source video and audio_path as the audio file; set mix to combine or false to replace.

Add procedurally generated audio to a video. One-shot convenience function to generate and add audio to video. Args: input_path: Absolute path to input video. audio_config: Configuration dict with: - drone: {"frequency", "volume"} for background tone - events: List of timed sound events output_path: Absolute path for output video. Returns: Dict with success status and output_path.

Overlay text on a video (titles, captions, watermarks). Returns JSON {success, output_path}. Use to burn a single text element into a video. Pass input_path as the source video, text with the string to draw, and position/size/color to style it; output_path is auto-named when omitted. Args: input_path: Absolute path to the input video. text: Text to overlay. position: Position on screen. Named (top-left, top-center, etc.), pixel" " {"x": 100, "y": 50}, or percentage {"x_pct": 0.5, "y_pct": 0.5}. font: Path to font file. Uses system default if omitted. size: Font size in pixels. color: Text color (CSS color name or hex). shadow: Add text shadow for readability. start_time: When the text appears (seconds). Null = always visible. duration: How long text is visible (seconds). Requires start_time. output_path: Where to save the output. Auto-generated if omitted. crf: Override CRF value (0-51, lower = better quality). Default 23. preset: Override FFmpeg encoding preset (ultrafast, fast, medium, slow, veryslow).

Overlay multiple text elements on a video in a single FFmpeg pass. Returns JSON {success, output_path}. Use when one video needs several captions at once. Pass input_path as the source video and texts as a list of overlay dicts (text plus optional position/size/color/timing). Args: input_path: Absolute path to the input video. texts: List of text overlay dicts. Each dict may contain: - text (str, required) - position (str|dict, default "center") - font (str, optional) - size (int, default 48) - color (str, default "white") - shadow (bool, default True) - start_time (float, optional) - duration (float, optional) output_path: Where to save the output. Auto-generated if omitted. crf: Override CRF value (0-51, lower = better quality). Default 23. preset: Override FFmpeg encoding preset (ultrafast, fast, medium, slow, veryslow). auto_layout: Automatically distribute vertically stacked texts at the same named position. Default True.

Apply a color grade to a video — by LUT file, style preset, or reference video. Args: input_path: Video file to grade. output_path: Where to write the graded video. reference_path: Optional reference video — when given, the video's color balance is adjusted to match the reference (overrides style). style: Style preset. One of: auto (gentle contrast lift), warm, cool, vintage, cinematic, dramatic, noir (high contrast, desaturated). lut_path: Optional .cube/.3dl LUT file applied with FFmpeg lut3d — overrides both reference and style for professional grading looks.

Remove silent sections from video.

Detect scene changes in video.

Separate audio into stems using Demucs.

Transcribe speech to text using Whisper.

Upscale video using AI super-resolution.

Comprehensive video analysis — transcript, metadata, scenes, audio, quality, chapters, colors. Accepts a local file path or an HTTP/HTTPS URL. Direct video URLs (e.g. https://example.com/clip.mp4) are downloaded automatically. Streaming-platform URLs (YouTube, Vimeo, TikTok, Twitter/X, Instagram, Twitch, …) require yt-dlp (pip install yt-dlp). Each sub-analysis is independent so one failure will not abort the others. Args: input_path: Local path or HTTP/HTTPS URL to the video. whisper_model: Whisper model size (tiny, base, small, medium, large, turbo). language: Language code for transcription (auto-detect if None). scene_threshold: Scene change sensitivity 0.0-1.0. include_transcript: Run speech-to-text via Whisper (requires openai-whisper). include_scenes: Detect scene changes and boundaries. include_audio: Analyse audio waveform, peaks, and silence regions. include_quality: Run visual quality check. include_chapters: Auto-generate chapter markers from scene changes. include_colors: Extract dominant colors and extended metadata. output_srt: Optional path to write SRT subtitle file. output_txt: Optional path to write plain-text transcript. output_md: Optional path to write Markdown transcript with timestamps. output_json: Optional path to write full JSON transcript data.

Apply an image mask to a video with edge feathering. Args: input_path: Absolute path to the input video. mask_path: Absolute path to the mask image (white = visible, black = transparent). feather: Feather/blur amount at mask edges in pixels (default 5). output_path: Where to save the output. Auto-generated if omitted.

Mix a looped music bed under a voice source with ducking and loudness normalization. The voice drives sidechain ducking and loudness normalization. A deterministic AudioBedReceipt is returned, and output duration stays locked to the voice source. Args: voice_source: Video/audio whose existing audio (voice/dialog) drives ducking and sets output duration. music_path: Background music or ambience to mix underneath; looped when shorter than the voice. output_path: Absolute path for the rendered output file. loop: Loop the music bed when it is shorter than the voice. Default True. loop_crossfade: Crossfade seconds between loop iterations. Default 1.5. fade_in: Output fade-in seconds. Default 0.0. fade_out: Output fade-out seconds. Default 2.2. target_lufs: Loudness normalization target in LUFS. Default -16.0. duck_threshold: Sidechain level above which ducking engages. Default 0.02. duck_ratio: Compression ratio applied while voice plays. Default 5.0. duck_attack: Duck attack in milliseconds. Default 25.0. duck_release: Duck release in milliseconds. Default 450.0. music_volume: Base music level before ducking. Default 1.0. save_receipt: Optional absolute path to write the AudioBedReceipt JSON. Returns: Dict with success status, output_path, output_duration, ducking_engaged, warnings, elapsed_ms, and the structured receipt.

Apply 3D spatial audio positioning.

Extract audio waveform data (peaks and silence regions). Args: input_path: Absolute path to the input video/audio file. bins: Number of time segments to analyze (default 50).

Audiogram + chapter mark plan (TE.4).

Auto-detect scene changes and create chapters. Analyzes video for scene cuts and returns chapter timestamps. Args: input_path: Absolute path to input video. threshold: Scene detection threshold 0-1. Default 0.3. Returns: List of (timestamp, description) chapter tuples.

Apply the same operation to multiple video files. Args: inputs: List of absolute paths to input video files. operation: Operation (trim, resize, convert, filter, blur, color_grade," " watermark, speed, fade, normalize_audio). params: Parameters for the operation. output_dir: Directory for output files. Auto-generated if omitted.

Run the AI-video benchmark corpus against host capabilities.

Replace video while preserving approved audio with explicit duration policy.

Save or load a brand kit / style profile JSON.

Probe host capabilities (ffmpeg, whisper, c2pa) and return a capability report.

Remove a solid color background (green screen / chroma key). Args: input_path: Absolute path to the input video. color: Color to make transparent in hex format (default green: 0x00FF00). similarity: How similar colors need to be to be keyed out (0.0-1.0, default 0.01). blend: How much to blend the keyed color (default 0.0). output_path: Where to save the output. Auto-generated if omitted.

Delete intermediate video files after a workflow. Useful for multi-step pipelines that leave temporary outputs. Files in ``keep`` are preserved even if listed in ``files``. Args: files: List of absolute paths to delete. keep: List of absolute paths to preserve (optional).

Compare video quality between original and processed versions. Args: original_path: Absolute path to the original/reference video. distorted_path: Absolute path to the processed/distorted video. metrics: Metrics to compute (default: ['psnr', 'ssim']).

Composite ordered image/video layers from a JSON spec. Supports explicit straight/premultiplied input alpha, opacity, positioning, transforms, timing windows, masks/mattes, allowlisted blend modes, rotation, and effect-noise routes to named layer/mask/mask-edge streams. Emits a deterministic layer-plan receipt; unsupported routes fail closed. File sources and masks must resolve inside the spec's directory. Args: spec_path: Path to a composite-layers JSON spec. output_path: Optional destination media path. save_layer_plan: Optional JSON path for the resolved layer-plan receipt. dry_run: Validate and emit the layer plan without rendering media.

Build source-backed manifests, selections, compositions, previews, and checks.

Transcode a video into a different container or codec (mp4, webm, gif, mov, ...). Returns JSON {success, output_path}. Use for format conversion; prefer video_export for delivery presets. Pass input_path as the source video and format as the target container; long renders stream MCP progress.

Sum known USD cost events by category for the project.