跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
视频与音频

timeless-love-story

Generates a ~70-second cinematic short where one couple is reincarnated across five historical eras plus a modern-day finale, shot in a consistent handheld DV-camcorder "found footage" style, scored with epic-romantic orchestral strings. Requires two reference photos (man, woman) and their two names — ask for these if missing. Use when the user asks to make a multi-era love story video, a reincarnated-lovers video, a "Timeless"-style short, or describes this exact concept (couple across history eras, camcorder POV, doorway transitions between eras). Always use this skill for this request rather than improvising the flow — the execution order (sheets first, then everything in parallel) is the point of the skill.

palmier-io · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add palmier-io/palmier-skills --skill timeless-love-story

技能说明


name: timeless-love-story description: Generates a ~70-second cinematic short where one couple is reincarnated across five historical eras plus a modern-day finale, shot in a consistent handheld DV-camcorder "found footage" style, scored with epic-romantic orchestral strings. Requires two reference photos (man, woman) and their two names — ask for these if missing. Use when the user asks to make a multi-era love story video, a reincarnated-lovers video, a "Timeless"-style short, or describes this exact concept (couple across history eras, camcorder POV, doorway transitions between eras). Always use this skill for this request rather than improvising the flow — the execution order (sheets first, then everything in parallel) is the point of the skill.

🎬 Timeless — Multi-Era Love Story Generator

What this makes

A ~70-second cinematic short where one couple is reincarnated across five historical eras plus a modern-day finale, all shot in a consistent handheld DV-camcorder "found footage" style. The story: in every era the two lovers find each other, share a beat of action/romance, and the man (always holding the camcorder, first-person POV) follows the woman through a dark doorway/threshold that transitions seamlessly into the NEXT era. Ends with a present-day subway reunion embrace and a script title card of their two names, scored with an epic-romantic orchestral strings track.

The eras, in order: Ancient Egypt → Ancient Rome → Medieval → Wild West → Renaissance → Modern Day (finale). (You may trim to fewer if the user wants a shorter piece, but this is the canonical set.) Medieval comes before Wild West — not alphabetical/chronological order — because the shot lists are written as a match-cut: Medieval ENDS on the couple running hand-in-hand through a dark stone gateway, and Wild West OPENS on that exact hand-in-hand run bursting out of a gateway into daylight. Swapping this order breaks the continuity the prompts were written for.

The ONLY things you need from the user

Before doing anything, you MUST have:

  1. Two reference images — one of the man, one of the woman (any photos; they anchor facial identity across every era).
  2. Two names — the man's and the woman's, for the closing title card.

If either is missing, STOP and ask for it. Example: "Send me a photo of each person and their two names, and I'll build the whole thing." Do not proceed on assumptions. You may optionally ask two refinements (but default sensibly if they don't answer): aspect ratio (default 16:9 horizontal — the character sheets are always horizontal regardless), and whether five eras is right.

Golden rules of execution order (this is the whole point of the skill)

  1. Session check first. get_timeline once. If canGenerate is false, tell the user to sign in to Palmier and subscribe — generation will fail otherwise. Read the project's actual fps from this response and use it for all frame math below — do not assume 24fps. (The reference build that produced the numbers in Step 3 happened to run at 24fps, 1280×720; a 30fps project needs every frame count recalculated, e.g. an 11s clip is 264 frames at 24fps but 330 frames at 30fps.)
  2. HARD GATE — photos before videos. Generate ALL character sheets first and do not call generate_video (or start any era clip) until every sheet is fully finished — get_media shows no generationStatus on each. Placeholders / still-generating sheets are not enough. Videos take those sheets as referenceImageMediaRefs; firing them early means broken or missing identity. Music may wait with the video batch, but never ahead of sheet readiness.
  3. Only after ALL sheets are ready, fire EVERYTHING else in parallel and don't block:
    • Generate all 6 era videos in one batch.
    • Generate the music track.
    • As each generation returns a placeholder id, immediately place it on the timeline even while it's still generating (Palmier allows placing generating clips — they fill in automatically when done). Add the text card too. This way the instant the last render finishes, the finished edit already exists — no assembly step afterward.
  4. Don't busy-poll. Fire generations, place placeholders, move on. Only read get_media ids:[...] when you genuinely need to confirm readiness.

MODEL CHOICES (resolve via list_models first — ids drift)

  • Character sheets (images): seedream-v5-pro (Seedream 5.0 Pro). It gave the most photoreal, consistent results in testing. Note its aspect ratios use different labels — 16:9 horizontal is landscape_16_9, NOT 16:9 (that errors).
  • Era videos: seedance-2 (Seedance 2, the regular/quality one — not Fast), 720p, duration 11s each (the finale is 14s). Pass the era's man+woman sheets as referenceImageMediaRefs and reference them in-prompt as @Image1 (man) and @Image2 (woman).
  • Music: elevenlabs-music, instrumental: true, duration 90 (closest supported value ≥ the ~70s runtime; you trim it down on the timeline).

STEP 1 — Character sheets (10 images: man + woman × 5 costume eras)

For the Modern finale, generate 2 more sheets (man + woman) — 12 total — but you can do those in the same batch. Every sheet uses this fixed deterministic 6-panel layout so they're consistent (this exact template mattered — free-form layouts came out inconsistent):

Fixed 2-row × 3-column grid. Top row: (1) full body front, (2) full body side profile, (3) close-up face front. Bottom row: (4) close-up face left profile, (5) close-up face back-of-head, (6) close-up face right profile.

Prompt template (fill {ERA COSTUME} per era, keep everything else verbatim):

Ultra-photorealistic DSLR photograph, single-character reference sheet, fixed 2-row by
3-column grid layout, 16:9, neutral seamless gray studio backdrop, consistent studio
strobe lighting across all 
阅读完整来源 ↗