Blocking to Video|场景调度与视频提示词

    1

    Plan spatially consistent AI-video shots from scripts and existing assets: blocking maps, camera directions, detailed performances, prop continuity, model-aware segments and targeted revisions.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Blocking to Video|场景调度与视频提示词

    Example session with this skill installed

    Use $blocking-to-video for the fictional reading-room scene below. A has just asked B, “这本书你看完了吗?” B replies “看完了。” then adds “封面。” Keep these exact lines. Plan the continuation with A west, B east, a closed book on the table, soft light from the west window, and all cameras on the south side of the dialogue axis. Reuse person-a.png, person-b.png and room.png; specify the two required control diagrams without claiming they already exist. My model has a 5-second minimum; other duration steps are unconfirmed. Write prompts only; do not generate media.

    • Read your context and instructions
    • Compiled the blocking to video
    • Generated the document

    Illustrative output — fictional planning example

    This is an original text demonstration, not a generated or visually verified film. The room layout is an assumption supplied in the request. The actual image and model node must be checked before generation.

    Reference plan

    FileResponsibilityStatus
    person-a.pngA identity and clothingUser-supplied; inspect before use
    person-b.pngB identity and clothingUser-supplied; inspect before use
    room.pngRoom architecture, furniture and light sourcesUser-supplied; inspect before use
    room-points.pngA west, B east, closed book at table P1Required control diagram; not yet created
    camera-map.pngC0 south, CA southeast toward A, CB southwest toward BRequired control diagram; not yet created

    Do not order additional character or environment images without a shot-specific reason. Draw control overlays locally when the scene image provides enough evidence. Do not render labels, dots or camera icons in the film.

    Segment and entry state

    A has finished the question and is looking at B with a closed mouth. Both sit in their established positions with empty hands. The book remains closed at P1. Editorial content is estimated at 3.6 seconds; this is below the stated 5-second model minimum. Use a supported duration of at least 5 seconds, confirm its availability in the current node, and trim the neutral tail rather than repeating a line or action.

    H02 — segment 0.0–1.4 seconds

    Medium close-up on B from CB, south of the axis, looking left toward offscreen A. Use B's actual room background as derived from the camera direction; do not substitute a generic exterior. B begins upright, shoulders relaxed. Say “看完了。” once, calmly, with one small nod during 0.1–1.1 seconds. Return the head to neutral; hands stay on the lap. A remains offscreen and silent. The camera is fixed.

    H03 — segment 1.4–2.6 seconds

    Cut to A from CA, still south of the axis. A looks right toward B. B says “封面。” offscreen during segment 1.6–2.3 seconds. A keeps the mouth closed while listening. Only after hearing “封面” do A's eyebrows rise and gaze stop. Preserve the west-window light in world space. No book movement and no extra camera pullback.

    H04 — segment 2.6–3.6 seconds

    Continue the same A shot without an unnecessary jump cut. A holds a brief blank stare, lips part slightly without speech, then remains still enough for the reaction to read. Both characters retain their positions and the closed book stays at P1. No added explanation or canned laughter.

    Tail, exit and checks

    If a 5-second generation is available, continue a quiet reaction hold to 5.0 seconds with no repeated dialogue, nod or prop action. Trim to the intended edit after review. Carry forward A's completed reaction, B's finished lines and the unchanged book state. Check exact dialogue order, on/offscreen speaker, camera side, background direction and visible prop state. Prompt timing is estimated; model execution, audio and visual continuity have not been validated by this text example.

    person-a.png

    PDF · document

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Blocking to Video — from confirmed script to usable shot prompts

    A production-workflow skill for creators who already have a story and want clear instructions for staging, camera placement, performance and continuity. It turns a confirmed script and available visual assets into a structured plan for AI-video generation, then helps revise that plan using the footage you actually accept.

    Designed for practical production problems

    • A character disappears from an establishing shot.
    • A reverse shot shows the wrong side of the room.
    • A fixed wide shot unexpectedly pulls back.
    • Expressions are vague or reactions happen before the important line.
    • Props change hands, duplicate or appear before they are obtained.
    • Too many scene assets are generated without a specific shot that uses them.
    • Short editorial shots do not match the video model's minimum generation duration.

    The skill provides checks and prompt structure for these issues. Model execution still needs to be reviewed; a prompt is not a guarantee.

    What the agent helps you produce

    1. Production status and asset map — adopted script, finished/accepted clips, next segment, available assets and specific missing items.
    2. Scene blocking and camera plan — character positions, sightlines, props, movement paths, camera directions and dialogue-axis side, grounded in the actual scene reference. Inferred space is marked as inferred.
    3. Reference responsibilities — real filenames and a clear purpose for every character, scene, prop and control image. Two-character scenes use a five-reference baseline; identity references scale with the cast.
    4. Detailed shot prompts — who must be visible, framing, background direction, start pose, triggers, facial changes, body mechanics, dialogue, listener reactions, prop contact, lighting, sound and the cut point.
    5. Generation-segment mapping — editorial shots grouped into durations supported by your current model, with explicit cut points and no automatic fast-forwarding or deleted dialogue.
    6. Continuity handoffs and targeted revisions — positions, gaze, held objects and completed actions carried between segments. Accepted footage is preserved; only affected instructions are revised.

    Detailed performance, props and lighting

    Emotions are translated into visible behavior: a trigger, two or three observable changes and an end state. Physical actions describe balance, support, contact and follow-through. Props have an owner, a count, a held hand, an entry point and a release state. Lighting follows the world-space sources in each scene, rather than applying the same tint or screen-side light everywhere.

    What is in the package

    • SKILL.md with stage selection and core workflow.
    • References for spatial assets, video prompts, direct-reference production, revision and validation.
    • A reusable segment-prompt template.
    • Original fictional text/JSON examples; example image filenames are placeholders, not included images.
    • A Python structural checker for asset mappings, dialogue, timing and camera/blocking fields.

    A practical way to use it

    1. Install the skill in your agent's supported skills directory and invoke $blocking-to-video.
    2. Provide your adopted script, actual asset files and current model settings. If clips already exist, say which are accepted and where to continue.
    3. Ask for the missing production step: asset audit, blocking/camera plan, shot prompts or a targeted repair.
    4. Bind the required references in your generation tool, generate with your own account/credits, review the result and feed back the specific failure.

    Example request: “Use $blocking-to-video. Keep my dialogue unchanged, reuse the available assets, map character positions and cameras, and write standalone generation segments with detailed reactions and prop continuity. Do not generate media yet.”

    Compatibility and requirements

    Works as instructions in agents supporting the SKILL.md standard. The agent needs access to the script and visual assets to inspect them, and file-writing capability when you request saved artifacts. Python 3.9+ is optional for the included checker. Your chosen video service must support the references and durations used in the plan; platform capabilities must be checked for the current node. No particular video provider or paid account is bundled.

    Scope, price and license

    US$9.90, one-time purchase for this downloadable workflow package. It does not include model credits, video rendering, custom production services, finished scenes, character image packs or guaranteed results. The structural checker does not verify identity, actual image geometry, lip sync, voices or final artistic quality. Audio reference binding and footage review remain separate checks.

    The current skill source is also publicly available on GitHub under the MIT license. This purchase is a convenient packaged distribution and a way to support its creator; it is not exclusive access or a different proprietary license. No lifetime-update or individual-support commitment is included in this description.

    Source and documentation: https://github.com/cgzs-li/blocking-to-video

    Does not generate video files or host assets. Optimal for two-character scenes. Requires user to provide initial character and environment master images.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 4 days ago

    • Free with an Agensi account

    Requires

    Windows
    Python
    Node.js

    Permissions

    Terminal / Shell
    Listed10 days ago
    Updated1 day ago

    What's inside

    Frequently Asked Questions

    It maps character positions, camera direction, eyelines and prop states, then carries explicit entry/exit states between segments. These instructions reduce ambiguity; they do not force a model to obey or guarantee error-free footage.

    Use an agent that supports SKILL.md. Video generation is separate: confirm the current model accepts the required references, durations and audio inputs. No particular model integration or model credits are included.

    The downloadable blocking-to-video skill folder contains the main instructions, spatial and video-production references, a reusable prompt template, original fictional text/JSON examples and a Python structural checker. Example photos, character packs, finished videos and custom production work are not included.

    Yes. Two characters are a baseline example, not a cast limit. Identity references scale with the cast, while each shot states who is present, offscreen or entering. Complex movement still requires inspection of the real scene and the generated result.

    A confirmed script is the starting point. With incomplete assets, ask for a minimum asset plan. With existing assets, reuse them and fill only shot-specific gaps. It does not automatically require new keyframes, multi-panel boards or duplicate environments.

    The current source is also public on GitHub under MIT. The paid listing offers a packaged distribution and supports the creator; it is not exclusive access. This listing does not promise lifetime updates, custom revisions or individual production support.

    New in Agents & Orchestration