one output05 / 14
Generate one clip
One clip from text, a first frame, first plus last frame, or references — priced before it runs.
apipod-generate-videoSKILL.md on GitHub
When to use it, and when not to
when to use it
- A single-clip request with a defined duration and aspect ratio.
- The user already has a still they want animated.
- A cheap draft is enough to judge the direction before a full-resolution take.
not for
- apipod-short-film — More than one shot, and continuity across cuts.
- apipod-vertical-video — The target is a 9:16 feed with its own safe-area rules.
procedure
5 skillsWhat the skill does, in order.
- 01
Price first — video is the expensive surface
Unless a cost was already approved, run estimate_cost and show the number before generate_video. Video is billed per second and the spread between models is wide.
- 02
Standardise on a still you can see
For new footage, make one keyframe at the video's aspect ratio, show it, and pass it as image_urls[0] on approval. A still the user has seen beats a prompt the model interprets — but say it adds one image generation.
- 03
Define the clip before prompting it
Purpose, aspect ratio, duration, subject, one primary action, one camera move, sound or dialogue, and how it ends. One action and one camera idea per clip.
- 04
Respect the duration enum
Read get_model_params for the durations and resolutions the model actually accepts. Rounding to a number you assume exists produces a rejection, not a clip.
- 05
Poll every 10 s and say so once
Video takes 1–5 minutes. Tell the user that once, then keep polling quietly. Deliver the URL, the final cost_usd, and a plain note on identity, motion, timing and audio sync.
try it
Send it from here
example request
Use APIPod to make a 5-second 720p video of ocean waves at sunset, with native audio.
The buttons open a new chat with this prompt prefilled. If APIPod is not connected yet, the prompt tells your assistant to stop and help you connect first. In Claude Code or a terminal, copy the prompt instead.
Not set up yet? Wiring guides for Claude, ChatGPT, Codex and more →
call trace
list_models (category: "video")
→ seedance listed
get_model_params (duration · generate_audio)
→ input_schema
estimate_cost (duration: "5", resolution: "720p")
→ $— · sufficient
generate_video (prompt written chronologically)
→ vid_7f3a9c
check_job (every 10 s)
→ completed · 1–5 min
Illustrative: the tools this skill runs, in order, for the request on the left. Prices appear the way estimate_cost returns them live — this page itself quotes nothing.
what it will not do
The rules it holds itself to.
01
One take per request. A revision is proposed, not run — it costs again and needs a yes.
02
Quote dialogue exactly and only when the model generates speech. Scene numbers and production notes get read aloud by speech models.
03
When the model exposes a draft or low-resolution option, offer it for the first take on long or expensive clips; finalise at full resolution only after the user approves the draft.