Skills

A clip from a picture you already have

A still becomes frame one and the model carries on from there.

Your agent reads what each video model charges per second before it generates anything, and runs the short version first so you can see the idea before paying for the long one.

You say: "Animate this product shot. Slow push in, six seconds, vertical."

Three steps to have your agent running this

  1. Connect Bildy at https://bildy.ai/mcp and sign in with your Bildy account. Setup per client.
  2. Copy the text below and save it as .claude/skills/animate-a-still/SKILL.md, or paste it into whatever your agent reads for instructions.
  3. Ask for what you want. Your agent reads the skill and follows it.
Download the file

Bildy tools this skill calls

  • files_upload
  • videos_list_models
  • videos_generate
  • videos_get_job

The whole file

---
name: bildy-animate-a-still
description: Turn a photo, a screenshot or a generated image into a short video clip, choosing a model that accepts a first frame and a clip length the budget allows. Use when someone wants to animate a picture they already have, add motion to a still, or make a short clip from artwork.
---

# A clip from a picture you already have

Someone has a still and wants it to move. Most video models take a first frame
and carry on from it, so the picture they give you becomes frame one and the
model invents the rest.

## Before you spend anything

Video costs much more than an image. Call `videos_list_models` first. That
call lists which models need a first frame and what each clip length costs.
Read the number for the length they asked for and say it before you generate.

If they want to see the idea first, run the shortest length the model allows,
show it, and only then run the long version.

## Steps

1. Get the picture as a link. A picture Bildy made already has one from
   `images_get_job`. A file on their computer gets one from `files_upload`:
   upload it with the command that call gives you and keep its `url`.
2. Pick a model from `videos_list_models` that accepts a first frame.
3. Call `videos_generate` with the picture's link in `imageUrl`, the prompt,
   the `ratio` that matches the picture, and `seconds`.
4. A clip takes minutes. Poll `videos_get_job` every fifteen seconds or so and
   tell the person it is still running rather than going quiet.
5. Hand over the download link when the status is completed. It expires after
   an hour.

## Writing a video prompt

Describe the motion. The model can already see the picture. "A slow push in
towards the bottle, the condensation running down, the background falling out
of focus" gives it something to do. "A bottle on a table" gives it nothing and
you get a still that wobbles.

Name the camera move in the words camera people use: push in, pull back, pan
left, tilt up, orbit, handheld. Models are trained on those.

Keep each clip to one action. Four seconds is not enough time for a person to
walk in, sit down and pick something up, and asking for all three gives you a
mess.

## What goes wrong

The face changes halfway through. Faces are the hardest thing to hold. Shorter
clips hold better, and a face taking up more of the frame holds better than a
face in the distance.

The clip drifts away from the picture after a second or two. That is the model
running out of the prompt. Add more to the prompt about what stays the same:
"the room stays exactly as it is, only the curtain moves".

Nothing moves at all. The prompt described the scene instead of the motion.

Other skills for the same tools