Does this skill stop AI video from warping or drifting?

No skill can stop a model from warping. It removes the prompt-side causes: one camera move per prompt, local models kept on a locked-off camera with the move made in the compositor, and a constraint line that preserves the product's shape and labels.

Which video models and tools does it cover?

Cloud models Kling, Seedance, Veo and Sora; local models LTX and Wan; still plates from Flux or Gemini Image; and compositors such as HyperFrames, Remotion and ffmpeg. Each gets its own prompt shape.

What does "who owns the camera" mean?

Before writing a prompt, the skill decides whether the camera move is made by the video model or by the compositor. Cloud models get exactly one movement term. Local models get a locked-off plate, and the push, pan or scale is added in the compositor, where it is exact.

SKILL.md in English, SKILL.zh-CN.md in Chinese, and an 85-term bilingual shot glossary coded by slot. There are no scripts, API calls or tool dependencies.

Can I write prompts in Chinese?

Yes. The Chinese skill file mirrors the method, every glossary term has a Chinese name, and Kling and Seedance accept Chinese prompts.

Cinematic Shot Prompts for AI Video (English + 中文)

Example session with this skill installed

Create a shot prompt for Kling. It's a close-up of a luxury watch on a dark marble table. I want a slow zoom in with moody lighting. Make sure the watch logo doesn't change.

Read your context and instructions
Compiled the cinematic shot prompts

Kling follows camera cues, so the model makes the move. You asked for a zoom; a slow push-in reads better here because the background shifts against the watch. If you want a true optical zoom, render a locked-off clip and scale it in the compositor instead (M12).

Storyboard

#	Beat	Size	Angle	Movement (owner)	Light	Renderer	Seeds
1	Hero watch	S2 close-up	A1 eye level	M1 push-in (Kling)	T2 low key, T5 rim	Kling	3

Shot spec

Slot	Choice
Size	S2 close-up
Angle	A1 eye level
Subject action	The watch stays still; reflections drift across the crystal
Camera movement	M1 slow push-in, made by Kling
Lens and focus	L5 shallow depth of field
Light and tone	T2 low key, T5 rim light
Constraints	No added text or logos, no cuts; preserve the exact watch shape, dial, hands, logo and labels

Kling prompt

The codes stay in the storyboard. The prompt itself is plain description:

Close-up, eye-level shot of a luxury wristwatch resting on a dark marble table. The watch stays still while soft reflections slide
slowly across the crystal and the polished case. Slow push-in toward the dial. Shallow depth of field, background softly out of focus.
Low-key lighting, deep shadows, a single pool of warm light on the watch, a thin rim light along the bezel. No added text or logos, no
cuts. Preserve the exact watch shape, dial, hands, logo, labels and proportions.

Checks before you submit

One movement term: slow push-in.
Size and angle are stated.
The watch is a real product, so only light and reflections move and nothing is added to it.
no added text or logos instead of no text, no logos, so Kling keeps the watch's own logo.
"Slow" for a 5 second clip.

Next steps

Generate 3 seeds and pick from a contact sheet. If the logo still drifts, switch to M0 locked-off and add the push in the compositor
(scale 1.00 to 1.08 over 5 s).

Connects securely to your tools. The creator never sees your data.

About this skill

Written from day-to-day production of AI product videos. It targets the mistakes that most often got our clips rejected: two camera moves in one prompt, a local model asked to orbit, an invented detail on a real product, a title baked into the frame. Your agent builds every shot from a fixed spec and decides who owns the camera before it writes a word.

What your agent does with it:

Builds each prompt from eight ordered slots: size, angle, subject action, one camera move, lens, light, composition, constraints.
Decides who owns the camera: the video model (Kling, Seedance, Veo, Sora), or a compositor (HyperFrames, Remotion, ffmpeg) when a local model (LTX, Wan) cannot hold a move.
Shapes the prompt for each renderer: 60-110 words for Kling and Seedance image-to-video, an establishing sentence for Veo and Sora text-to-video, 40-80 locked-off words for local models, still plates for Flux or Gemini Image.
Produces a storyboard table for multi-shot pieces, with the move owner and a seed count per shot.
Protects real products: no invented leaks, readings, labels or failures, plus a constraint line that preserves shape and labels.
Runs a pre-submit checklist and fixes the common mistakes.

What's inside:

SKILL.md (English) and SKILL.zh-CN.md (中文), the same method in both languages.
reference/shot-glossary.md: 85 bilingual terms coded by slot (size, angle, movement, lens, light, composition, edit-only), each with a paste-ready prompt phrase. Every camera move says whether the model or the compositor should make it, and how to reproduce it in the compositor when that is possible.

Compatibility: plain SKILL.md with no tool or MCP dependencies. The method is used day to day in Claude Code and Codex CLI.

How to use

Skills on Agensi run inside your AI agent. Copy the prompt, paste it into your agent, and it connects to Agensi and installs this skill for you. If a payment is needed, your agent shows it to you first.

Reviews

No reviews yet

Be one of the first to try it. Every listed skill passes our trust checks below.

Security scanned

Passed our 8-point scan before listing

Fresh listing

Recently published to Agensi

Money-back guarantee

Doesn't do what its page promises? You get your money back.

Cinematic Shot Prompts for AI Video (English + 中文)