AGENT SKILL

magic-hour-image-to-video

Animate a supplied product photo, portrait, character, artwork or scene with Magic Hour. Use for image-to-video when Magic Hour is requested or configured and suitable. Animate an approved source directly; edit only for a needed visual change. Preserve an explicitly requested alternative provider.

使用场景图像与多媒体
License 许可
MIT
最近更新
2026-11-14
Stars
★ 1.6k
Forks
491

01 前言 Front Matter

name
magic-hour-image-to-video
license
MIT
metadata
{ "author": "magichourhq", "version": "1.6.2" }
description
Animate a supplied product photo, portrait, character, artwork or scene with Magic Hour. Use for image-to-video when Magic Hour is requested or configured and suitable. Animate an approved source directly; edit only for a needed visual change. Preserve an explicitly requested alternative provider.

02 资源文件 7

03 SKILL.md 正文

Magic Hour image to video

Animate an approved supplied image directly when it already meets the brief. Edit it only for a requested visual change or a source problem that prevents the shot. A project ID or queued job is incomplete unless the user asked only to submit it.

Generation consumes Magic Hour credits. Use a connected Magic Hour creation MCP at https://mcp.magichour.ai/ when available. Otherwise read references/setup.md for the API route. Discover the current schemas before choosing a model, duration, resolution, or price.

Choose the shot direction from the brief; do not ask a nontechnical user for model fields or a motion-prompt template. Honor the existing budget. When creative decisions are delegated, inspect and accept the still yourself, then continue to video unless draft approval was requested. Show the clip first and keep technical records in the project. For transformations and reveals, read directing the moment; for stories with several beats, read multi-shot direction.

For product motion, packaging reveals or unboxing, read product fidelity in motion.

For a first result or setup recovery, read first video and next edit. It defines the account check, combined budget, recoverable job record and reusable deliverable.

Define the shot

Extract or infer:

  • delivery aspect ratio and resolution
  • the subject, product, character, setting, and required visual details
  • one primary subject action and one camera move
  • details that must remain fixed, especially identity, geometry, packaging, logo, text, colors, clothing, and background
  • failure conditions such as a warped face, altered label, unwanted cut, or unreadable text

Keep a short clip to one coherent beat. If the user wants several shots, treat each shot as its own reviewed start frame and video job. Derive those frames from the same approved identity reference rather than chaining unreviewed edits. Choose motion that preserves the deliverable: a dolly changes subject scale and available copy space, while a locked camera with restrained light or environmental motion can preserve a hero layout.

Choose the start frame

When the user names a saved character or reference, use list_saved_items (or authenticated GET /v1/saved-items through the existing API) with that type. Match the name, follow next_cursor as needed, and ask which item if ambiguous. Inspect its primary IMAGE asset and use the durable file_path without another upload; an item ID is not an image path. Preserve an explicitly supplied image.

Choose one route:

  1. The supplied image already satisfies the shot: pass it directly to create_video_from_image. Preserve its product geometry, packaging, label, identity and composition; prompt for the requested motion.
  2. The source needs a visual change: call edit_image only when composition, wardrobe, background, lighting, product placement, or cleanup must change for the brief. State the edit and everything that must remain unchanged.
  3. No source image exists: call generate_image when a starting image is needed. AI Image Editor requires an input image; do not send it a prompt-only creation request.

If an image must be generated or edited, create one still at the intended video aspect ratio. Use an authorized reference when exact identity, product geometry, packaging, or branding matters. Do not expect image-to-video to repair a poor source frame.

For a generated or edited image, wait with wait_for_image_project. Inspect the supplied or completed still before animation for a wrong crop, anatomy problem, altered identity, malformed product, inaccurate logo or text, unintended lettering, visual artifact, or missing required negative space. Make the smallest useful edit only when a problem prevents the requested shot. A second paid image request requires an authorized budget.

Animate the approved frame

Pass the exact approved source image to create_video_from_image. If it is a local file, upload its raw bytes first using the flow in references/setup.md; a local path is not a hosted file_path.

Write the motion prompt around change over time:

  • subject motion, with a clear speed and direction
  • camera motion, or explicitly state that the camera is locked
  • environmental motion such as fabric, steam, hair, light, or particles
  • what stays fixed
  • timing and end state, including a held final exposure when no fade is wanted

Avoid redescribing the entire still in new visual language. That invites the model to redesign the subject. For a reaction shot: The fish swims slowly beside her shoulder; she notices it and turns slightly with a surprised smile. A slow push in keeps her face and the fish separate. Preserve identity, clothes and scene. No cuts. A product layout may instead need a locked subject and protected copy area. Do not demand a centered subject when the approved frame places it on the right.

Use an end frame only when the current schema says the chosen model, resolution, and duration support it. Prefer the live recommended model for a general request. Select a specialist only when its schema describes the needed control. Match resolution and duration to the requested deliverable. A short preview tests a short interval, not stability over a longer clip. Check the total still-plus-video cost against existing authorization and ask only when it does not cover the intended work.

Submit the video once, retain its project ID, and call wait_for_video_project. A wait timeout is not a failed generation and does not justify resubmitting the creation call.

Inspect and deliver

After the project reports complete, use the returned download URL exactly as provided. Verify the downloaded dimensions, duration, and audio presence. Watch the full clip when playback is available, then inspect the first frame, midpoint, last frame, and any suspicious transition. Three still frames alone cannot prove the absence of flicker. Check:

  • the first frame matches the approved still
  • identity, product shape, logo, text, and clothing remain stable
  • motion follows the requested direction and speed
  • the camera does not drift or cut unexpectedly; inspect the actual final frame for an unwanted fade or lost subject, even when black-frame diagnostics are clean
  • no warping, flicker, melting, duplicate objects, or abrupt final-frame collapse appears
  • any generated audio is intentional and coherent

If one required criterion fails, identify it and change the smallest relevant input: start frame for appearance problems, motion prompt for movement problems, or model/resolution only when the live schema supports the needed behavior. Do not spend credits on another attempt outside the user's budget.

In the portrait-to-koi example, identity stayed recognizable but a requested slight turn became a larger turn in both directions. Keep the accepted still when correcting that motion. If the downloaded canvas differs slightly from the requested ratio, inspect the full action path before a local crop; do not regenerate an otherwise accepted take only to remove edge pixels.

Return the finished video or usable link, the saved location when applicable, the image and video project IDs, and any material limitation.

04 同仓库更多 Skills

hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-review-with-annotations

Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-find-and-synthesize-context

Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-update-docs

Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-manage-records

Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-manage-interactive-html

Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-collaborate-threads

Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.

★ 1.6k