magic-hour-character-replace
Replace a subject in an existing video or transfer its performance onto a reference character with Magic Hour. Use for whole-character recasting and motion transfer; not face-only swaps or generating unrelated scenes with a recurring character.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 1.6k
- Forks
- 491
01 前言 Front Matter
- name
- magic-hour-character-replace
- license
- MIT
- metadata
- { "author": "magichourhq", "version": "1.2.0" }
- description
- Replace a subject in an existing video or transfer its performance onto a reference character with Magic Hour. Use for whole-character recasting and motion transfer; not face-only swaps or generating unrelated scenes with a recurring character.
02 资源文件 2
- 参考文档agents/openai.yaml274 B
- 许可证LICENSE1.1 KB
03 SKILL.md 正文
Magic Hour character replace
Reuse a performance the user already likes. Accept “make my character do this” or “replace this performer”; infer the correct mode from what the user wants to preserve. Keep the reference and source footage distinct and complete the requested clip within the existing budget.
Choose replace or animate
| Request | Mode | What to inspect |
|---|---|---|
| Put my character into this footage, keeping the setting | replace |
Intended subject replaced, source action and surrounding scene retained |
| Make the character in my image perform this motion | animate |
Reference character and image composition, transferred performance |
| Change only this person's face | Face Swap | Facial identity; body and costume are outside that operation |
| Same character in a newly invented scene | Image Editor → Image-to-Video | New approved still before motion |
Do not use a text prompt to invent a source performance that was already supplied. The current Character Replace schema has mode and subject-selection controls, not a general scene-prompt field. Do not fabricate camera, seed or strength settings.
Prepare the performance
Inspect the actual video interval, dimensions, cuts, subject count, visible limbs, occlusions and motion speed. Match the character reference to the visible body and framing when possible. A portrait cannot establish faithful unseen legs; complicated hand contacts and fast turns need a representative output check. Keep the original, rather than repeatedly feeding generated revisions back into the model.
For a named saved character, use list_saved_items (or authenticated GET /v1/saved-items through the existing API) with type: character, match the name and follow next_cursor as needed. Disambiguate multiple matches, inspect the primary IMAGE asset for performance-compatible framing, and pass its durable file_path as assets.image_file_path. Do not upload it again or pass the item ID. Preserve an explicitly supplied reference.
Use authorized footage and reference images. For a difficult or long source, generate a short representative interval before committing the remaining budget. Select the difficult action, not only the easiest stationary moment. Preserve any user-requested duration and review point.
For one unambiguous subject, use style.selection_mode: auto. For a specific subject in a crowd, use point and inspect a source frame to obtain position_x, position_y in original video pixels plus time_seconds. Coordinates from a resized preview must be scaled back. Time uses the source clock, including when start_seconds is nonzero. Never guess coordinates or add a cropped-clip offset twice. Ask only when the target person is ambiguous.
Execute and compare
Use replace_character_in_video with assets.image_file_path, assets.video_file_path, the source interval, supported resolution and style.mode. Use the creation MCP at https://mcp.magichour.ai/; setup. API users should consult the live schema for /v1/character-replace. Check final interval cost against the existing authorization.
Upload local inputs through get_upload_urls, PUT raw bytes without the API bearer token, then pass file_path. Create once, save the project ID, and use wait_for_video_project. Resume the same project on a timeout. Keep returned download URLs intact and save the completed file.
Review the whole result beside both inputs: correct target, recognizable character, intended mode, head/body proportions, hands and feet, source gesture timing, contact with objects, occlusion boundaries, background and audio. Multiple attractive frames cannot establish a preserved performance. Check every cut; a changed face or extra limb is a failed requirement.
Measure the actual output interval and dimensions before assembly. Our replace-mode example kept a recognizable fox character but returned 5.375 seconds for a 0–5-second request. Compare the same source moments; do not assume matching frame numbers or silently retime a required performance.
The animate-mode example uses the same reference and performance but keeps the reference's studio setting, while replace mode keeps the source station. Use that difference to honor “keep my character's world” versus “put my character in this footage.” Both require timing and full-playback review; one simple head turn does not establish dance or complex-contact quality.
If the wrong person changes, fix subject selection. If body shape fails, improve framing/reference compatibility or simplify the source interval. If the scene changes unexpectedly, verify the mode before spending again. Repair only the failed shot within budget. Do not describe all motion as faithfully transferred merely because the job completed.
Deliver the finished clip, source/reference/mode retained in the existing project brief, and unresolved defects. Reuse the same reference for the next performance; reuse the accepted performance when only the character changes. Retain the selected saved-item ID and durable asset path privately for the next performance; refresh previews through the library rather than regenerating the character.
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.