Usage Rules
- First read
../byted-mediakit-shared/SKILL.md, and perform the unified pre-check; if this Skill is missing, stop and prompt for installation. - Only select from the table below
videodomain tools; similar capabilities are distinguished by each tool's "capability description" and parameter boundaries. - Before execution, read the corresponding reference as needed; the parameter and result descriptions come from the same reviewed copy, and the complete machine contract is subject to the current CLI
--schema. - When required parameters, authentication environment variables, or real input resources are missing, ask the user for them; generic optional fields may only pass through values explicitly provided by the user, while other optional fields may be accurately determined from clear intent, but must not be fabricated.
- At execution time, set
MEDIAKIT_SURFACE=skill, specifying that the invocation source is Skill. - At execution time, set
MEDIAKIT_RUNTIMEto the current Agent host, to avoid the CLI being unable to reliably identify the parent Agent.
Clarification and Cross-Domain Routing
If it only says there is a video without specifying the business goal, clarification should be made first. Finished-video editing requests that clearly involve burning subtitles or images onto the final video, or performing cropping, concatenation, mixing, merging, speed adjustment, transitions, image rotation and flipping, video filters, or multi-video compositing should be routed to editing; single-image processing should be routed to image; audio transcoding, vocal separation, or voice activity detection should be routed to audio.
Tool List
| Tool | Description | Supported Modes | Command | Reference |
|---|---|---|---|---|
| add-video-invisible-watermark | Used for adding video dark watermarks. Without affecting the visual quality and integrity of the video image, a string of digital information is covertly embedded into the video file. It is suitable for scenarios such as video copyright protection, content leak tracing, and file authenticity verification. | Cloud | mediakit-cli video add-video-invisible-watermark |
reference/add-video-invisible-watermark.md |
| analyze-video-highlights | Supports two analysis models, short drama Miniseries and mini-game Game, for highlight clip extraction, and outputs precise timestamps, highlight scores, OCR text, and image descriptions for secondary development or content analysis. | Cloud | mediakit-cli video analyze-video-highlights |
reference/analyze-video-highlights.md |
| analyze-video-storyline | Used for plot storyline analysis, analyzing one or more long videos based on large-model video understanding and generating structured plot data. The analysis results include two parts: plot segments arranged in chronological order, and a highlight storyline organized and summarized based on the video segments. | Cloud | mediakit-cli video analyze-video-storyline |
reference/analyze-video-storyline.md |
| asr-subtitles | Recognizes and extracts subtitle text with timestamps from the speech in video or audio; suitable for requests such as extracting video subtitles, speech-to-subtitle, and dictating dialogue. The recognition target is the speech content in the audio track, not hard subtitles already burned into the image. | Cloud | mediakit-cli video asr-subtitles |
reference/asr-subtitles.md |
| assess-video-quality | Used for video image quality detection. | Cloud | mediakit-cli video assess-video-quality |
reference/assess-video-quality.md |
| drama-recap | Based on a completed script restoration task, custom narration or AI-generated narration can be used to generate marketing or narration videos with AI voiceover and narration subtitles; voice timbre, subtitle style, and original subtitle erasure can be configured. | Cloud | mediakit-cli video drama-recap |
reference/drama-recap.md |
| drama-recap-vertical | Based on understanding the characters and storyline of the input short drama episodes, automatically extract highlight clips and generate brand-new commentary videos; supports text commentary (original highlight mashup + on-screen text) and voiceover commentary (original highlight mashup + AI voice + BGM), and can apply the three-element visual template for short dramas. | Cloud | mediakit-cli video drama-recap-vertical |
reference/drama-recap-vertical.md |
| drama-script | Based on large model video understanding capabilities, convert short drama videos into structured script text, identifying and extracting core elements such as scenes, characters, dialogue, and plot. | Cloud | mediakit-cli video drama-script |
reference/drama-script.md |
| enhance-video | Used for video quality enhancement. Utilizes AI algorithms to analyze the input video and intelligently perform a series of optimization operations including but not limited to video denoising, color enhancement, clarity improvement, defect repair, and super-resolution. Provides two versions, standard and professional: standard balances processing speed and video quality, with more than 10 commonly used enhancement algorithms built in, suitable for quality enhancement in video distribution scenarios; professional provides ultimate quality enhancement, with more than 30 deep AI enhancement algorithms built in, suitable for film-grade video production. Different versions affect the intensity of enhancement algorithms, applicable scenarios, and billing. | Cloud | mediakit-cli video enhance-video |
reference/enhance-video.md |
| enhance-video-fast | Integrates lightweight super-resolution and intelligent quality enhancement, adopting a speed-first strategy to efficiently balance processing efficiency and image quality, especially suitable for latency-sensitive business scenarios. | Cloud | mediakit-cli video enhance-video-fast |
reference/enhance-video-fast.md |
| enhance-video-generative | Provides generative video enhancement and restoration based on Diffusion large model technology. Through deep semantic understanding, it intelligently completes and generates realistic details consistent with the video content, and can repair pixels lost during video compression or aging, ultimately producing natural, high-fidelity video images. | Cloud | mediakit-cli video enhance-video-generative |
reference/enhance-video-generative.md |
| erase-video-subtitle | Intelligently detects and erases existing hard subtitles in video frames while preserving the original background. Supported formats: mainstream video formats such as mp4, flv, ts, avi, mov, wmv, mkv. | Cloud | mediakit-cli video erase-video-subtitle |
reference/erase-video-subtitle.md |
| erase-video-subtitle-pro | Used for subtitle erasure (refined version), performing high-quality traceless erasure of video subtitles and restoring the video image to the greatest extent possible. | Cloud | mediakit-cli video erase-video-subtitle-pro |
reference/erase-video-subtitle-pro.md |
| extract-frames | Extracts screenshots from videos. The screenshot results can be used for video covers, preview images, sprite sheets, or as input for other video understanding tasks. | Cloud | mediakit-cli video extract-frames |
reference/extract-frames.md |
| extract-video-invisible-watermark | Parses and restores hidden digital information from videos embedded with invisible watermarks; if the same video has been embedded with invisible watermarks multiple times, all watermark information can also be extracted. | Cloud | mediakit-cli video extract-video-invisible-watermark |
reference/extract-video-invisible-watermark.md |
| face-blur-video | Video face masking can automatically and accurately identify face regions in video frames and apply blurring or mosaic processing to all faces, suitable for scenarios requiring protection of facial privacy. | Cloud | mediakit-cli video face-blur-video |
reference/face-blur-video.md |
| face-swap-video | Fuses and replaces the target face provided by the user onto the person in the video, outputting a high-quality face-swapped video, mainly suitable for scenarios requiring face swapping for generative video anonymization. | Cloud | mediakit-cli video face-swap-video |
reference/face-swap-video.md |
| generate-highlights-microdrama | Can be used for intelligent highlight editing of short dramas. Based on understanding the characters and storyline of the input episodes, it extracts highlight clips and generates videos such as highlight mashups and single-episode trailers according to requirements such as duration, number of outputs, sequential editing, or jump editing. | Cloud | mediakit-cli video generate-highlights-microdrama |
reference/generate-highlights-microdrama.md |
| generate-highlights-minigame | Supports identifying core gameplay and highlight events in mini-game screen recording videos, such as combos, level clearing, and extreme operations, and quickly generates video materials for user acquisition promotion. Optionally provides game name, gameplay description, and highlight definitions to assist in more accurately identifying exciting content. | Cloud | mediakit-cli video generate-highlights-minigame |
reference/generate-highlights-minigame.md |
| generate-highlights-movie | Supports long-form video content such as movies and TV series, identifying highlights along the plot storyline and splitting them into multiple highlight clips of specified duration, used as short video materials for film and television collection distribution; the algorithm identifies and removes low-density information such as scenery setup, slow camera movement, and opening and ending songs; each split segment features a highlight-front opening and an ending hook design. | Cloud | mediakit-cli video generate-highlights-movie |
reference/generate-highlights-movie.md |
| martencode-video | When transcoding, Ultra HD intelligently analyzes the scenes, actions, content, and textures of the video, selects the optimal encoding parameters, and outputs video with better subjective image quality at a relatively lower bitrate, reducing bandwidth costs and improving the user visual experience. | Cloud | mediakit-cli video martencode-video |
reference/martencode-video.md |
| matte-greenscreen-video | It can perform matting on videos with a green screen or solid-color background, automatically identify and retain the subject, and ultimately generate a video with a transparent or solid-color background. | Cloud / Local | mediakit-cli video matte-greenscreen-video |
reference/matte-greenscreen-video.md |
| matte-portrait-video | Automatically identifies the human subject in the video, removes the original background, and generates a video file with a transparent or solid-color background, suitable for post-processing scenarios such as background replacement. | Cloud | mediakit-cli video matte-portrait-video |
reference/matte-portrait-video.md |
| probe-video-metadata | Detects the input video URL and outputs standardized media asset metadata. | Cloud / Local | mediakit-cli video probe-video-metadata |
reference/probe-video-metadata.md |
| remux-video | Video remuxing is used to adjust the video container format. It only modifies the container format and does not re-encode or re-decode the audio and video streams. It is suitable for scenarios such as on-demand distribution adaptation, streaming media segmentation and packaging, and multi-platform compatibility. | Cloud | mediakit-cli video remux-video |
reference/remux-video.md |
| segment-scenes | Automatically splits multiple scene segments based on the video's transitions and changes in picture content, and outputs the timeline information of each scene segment and the corresponding independent video file. | Cloud | mediakit-cli video segment-scenes |
reference/segment-scenes.md |
| semantic-segment | Comprehensively analyzes the video's visuals, speech, and narrative structure, and intelligently splits a long video into multiple independent material clips through strategies such as shot change and speech pause detection, while ensuring semantic completeness and avoiding cutting a single sentence in the middle. | Cloud | mediakit-cli video semantic-segment |
reference/semantic-segment.md |
| transcode-video | Video transcoding converts a video stream into another video stream, and may involve conversion of encoding format, resolution, bitrate, I-frame interval, and container format. It is used to adapt to different business scenarios, playback terminals, and network environments. | Cloud | mediakit-cli video transcode-video |
reference/transcode-video.md |
| video-ocr | Used for video subtitle recognition (OCR), it recognizes subtitle information in the input video picture and outputs structured text data with timestamps. | Cloud | mediakit-cli video video-ocr |
reference/video-ocr.md |
| video-understand-router | Based on a large visual model, it performs general video content analysis on the input list of video URLs and outputs structured video-level understanding results, suitable for scenarios such as content moderation, video retrieval, and tag generation. | Cloud | mediakit-cli video video-understand-router |
reference/video-understand-router.md |