본문으로 건너뛰기
로그인무료로 시작

그림으로 짧은 영상

길이, 움직임, 가격을 먼저 정하고 그림을 짧은 영상으로 만들어요.

쓰는 법

어시스턴트의 스킬 폴더에 스킬을 복사하세요. Claude Code라면 ~/.claude/skills/<이름>/SKILL.md예요:

mkdir -p ~/.claude/skills/pickshot-clip-from-picture
curl -L https://pickshot.ai/developers/skills/pickshot-clip-from-picture/SKILL.md \
  -o ~/.claude/skills/pickshot-clip-from-picture/SKILL.md

SKILL.md 다운로드 · MCP 연결 방법 · 모든 스킬

SKILL.md 전체

---
name: pickshot-clip-from-picture
description: Turn a picture into a short video clip on the user's Pickshot account, with the length, shape and motion the user asks for and the price shown first. Use it when the user wants a picture to move, a product turned into a short clip, or a character shot animated.
---

# A clip from a picture

This skill needs the Pickshot MCP server (`get_balance`, `list_models`, `estimate_cost`, `create_video`, `get_job`). If they are missing, tell the user to connect it first (https://pickshot.ai/en/developers/mcp).

The spending rules of the `pickshot-generate` skill apply and matter most here, because clips cost the most: quote first, wait for yes, cap the price, one key per request, and never start a second clip while the first is running.

## Steps

1. Get the picture as a public https link. If the user wants one made first, use the `pickshot-generate` skill, then come back with its link.
2. Ask what should move and how the camera should move, in one or two sentences ("the steam rises, slow push in"). If they do not know, suggest one simple motion.
3. Call `list_models` with `kind` `video`. Keep models that are `available` and have the `i2v` mode. Read the limits for `i2v`: the lengths in `seconds`, the `resolutions`, the `ratios`, and whether `sound` is offered.
4. Prefer the cheapest model that fits. Choose the shortest length that tells the story (five seconds is usually enough). Do not turn `sound` on unless the user wants it.
5. Build the request: `model`, `mode` `i2v`, `prompt` (the motion), `image_url`, `seconds`, and `ratio` or `resolution` if the limits ask for them.
6. Call `estimate_cost` with `kind` `videos`. Tell the user the model, the length and the credits, and wait for yes. If a longer clip or a higher resolution is much dearer, say how much dearer.
7. Call `create_video` with the same request, `max_credits` equal to the quote and a new `idempotency_key`.
8. The answer is a job. Call `get_job` every 15 to 30 seconds until it is `done` or `failed`. Clips take minutes. Tell the user it is running, and do not create another one.
9. Give the user the link in `results[0].url`. If it failed, say so and ask before trying again.

## Prompt tips

- Describe motion, not the picture again: the picture already shows what is there.
- One subject, one movement, one camera move. Several ideas in one clip usually look confused.
- If the clip came out wrong, change the motion sentence, not the picture, and ask before paying again.