magic-hour-text-to-video
Turn an idea into a short Magic Hour video when no exact identity or starting frame is required. Use for original surreal scenes, visual jokes, atmosphere or concept exploration; use Image-to-Video for an approved person, character or composition.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 1.6k
- Forks
- 491
01 前言 Front Matter
- name
- magic-hour-text-to-video
- license
- MIT
- metadata
- { "author": "magichourhq", "version": "1.0.0" }
- description
- Turn an idea into a short Magic Hour video when no exact identity or starting frame is required. Use for original surreal scenes, visual jokes, atmosphere or concept exploration; use Image-to-Video for an approved person, character or composition.
02 资源文件 2
- 参考文档agents/openai.yaml255 B
- 许可证LICENSE1.1 KB
03 SKILL.md 正文
Magic Hour text to video
Translate the idea into one clear visual event. Accept “make an elevator open onto the moon”; choose the framing, pacing and model within the user's existing budget. Preserve requested creative choices. Deliver the video, with any unmet requirement made explicit.
Decide whether a reference is necessary
Use create_video_from_text for an original scene whose exact cast and appearance can vary. If the user provides their face, a recurring character or a required starting composition, use that reference through AI Image Editor when needed, then Image-to-Video. An optional imagined character is not a reason to force a paid image step. Honor an explicit Text-to-Video request while explaining any identity limit that matters.
Direct a visible event
Define the setup, the change and the payoff. For a short clip, choose one event that a viewer can understand without explanatory copy. Establish the subject immediately; reserve enough time to see the result. Give physical direction and timing instead of piling up adjectives.
Examples of structures, not measured viral formats:
| Intent | Shot direction |
|---|---|
| Impossible reveal | Familiar closed door → it opens onto an unexpected world → hold long enough to read the reveal |
| Visual joke | Establish an ordinary action → one clear physical contradiction → visible reaction or consequence |
| Cinematic atmosphere | One focal subject, one camera move and restrained environmental motion; no unnecessary plot |
Write subject/action → setting → framing/camera → timing/end state → essential constraints. Match the proposed action to the supported duration. Multiple unrelated events in five seconds can lose the payoff; use separately directed shots and an available compositor when the brief needs them. Avoid arbitrary fixed shot counts or slowing the whole video to hide missing action.
Example: “Locked camera inside a steel elevator. Its doors open during the first two seconds onto a lunar landscape with Earth above the horizon. Hold the impossible view for the remaining three seconds. Realistic cold light, straight door geometry, no people, text or cuts.”
Generate, review, repair
Use the creation MCP at https://mcp.magichour.ai/; setup. Read current model, duration, resolution, audio and aspect support. API users should consult the live schema for /v1/text-to-video. Set audio intentionally; an audio toggle does not guarantee exact dialogue, music rights or synchronized sound.
Submit one job within the existing spending authority, retain its ID, then use wait_for_video_project. Resume that ID after a wait timeout; never repeat an ambiguous create request. Download using the exact returned URL and verify actual dimensions, duration and streams.
Review the complete event: does the setup read, does the requested action happen, is the payoff visible, does geometry survive, and are cuts or sounds intentional? Judge physical interactions and timing, not just attractive still frames. Do not call a requested loop seamless without comparing the join.
Treat exclusions as requirements to verify. The LTX-2.5 moon-elevator run opens the doors, but they keep moving through most of the clip and the requested blue Earth is unclear. Check when the action finishes, what is visible, and how long the shot holds.
For an absent action, simplify the beat or provide clearer timing before changing everything at once. For wrong appearance, use an accepted still and Image-to-Video on the next authorized attempt. Preserve successful takes; do not repeatedly regenerate an accepted scene merely to change captions or export shape.
Deliver the finished file, the prompt and accepted settings in the project, and material limitations. For a series requiring the same cast, retain a usable frame as a reference for later scenes. Do not claim a text seed alone locks identity.
04 同仓库更多 Skills
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.
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.
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.
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.
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.
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.