video-prompting

The saved excerpt is a snapshot from when this skill was listed. The external source remains the complete and most current version.

---
name: video-prompting
description: Draft and refine prompts for video generation models (including text-to-video, image/keyframe-to-video, and reference-driven generation), and create character-sheet prompts for image models when the goal is character consistency before image-to-video. Use when a user asks for a "video prompt", a model-specific prompt such as MiniMax H3, Seedance 2.0, Seedance 2.5, Ovi, Veo 3, Wan 2.2, Wan Animate 2, LTX-2, LTX-2.3, or LTX-2.5, or a consistent-character prompt such as "character sheet prompt", "character turnaround", "character reference sheet", or "photographic identity sheet".
---

# Video Prompting

## Overview

Turn a user’s intent into either:

- a strong, model-compliant video prompt, or
- a strong image-model prompt for a character sheet that will later support image-to-video consistency.

Model-specific video guidance lives in `references/models/`. Character-sheet guidance lives in `references/workflows/character-sheets.md`.
This file is the entry point: route to the right path, ask the minimum clarifying questions, then draft the prompt in the expected format.

## Model Index

- Ovi: `references/models/ovi/prompting.md`
- Veo 3 / 3.1: `references/models/veo3/prompting.md`
- Wan 2.2: `references/models/wan22/prompting.md`
- Wan Animate 2: `references/models/wan-animate-2/prompting.md`
- Seedance 2.0: `references/models/seedance2/prompting.md`
- Seedance 2.5: `references/models/seedance2-5/prompting.md`
- MiniMax H3: `references/models/minimax-h3/prompting.md`
- LTX-2: `references/models/ltx2/prompting.md`
- LTX-2.3: `references/models/ltx2-3/prompting.md`
- LTX-2.5: `references/models/ltx2-5/prompting.md`

## Workflow Index

- Character sheets for consistent characters: `references/workflows/character-sheets.md`

To add a new model later: create `references/models/<model>/prompting.md`, then add it to this index.

To add a new workflow later: create `references/workflows/<workflow>.md`, then add it to the Workflow Index.

## Global video-prompt rules

These rules apply to every video model reference:

- Never include the model name, model version, duration, aspect ratio, resolution, or API/control parameter names in the final prompt text unless the selected model's required prompt schema explicitly includes timing. MiniMax H3 alignment instructions and shot timestamps are such an exception.
- Otherwise, use duration only as internal planning context for how many action beats the prompt can support.
- If the user asks for parameters or the model requires them, provide them outside the prompt in a separate recommended-parameters line.
- For image-to-video, treat the image as the visual anchor. Do not describe the image in depth unless the user asks for an image analysis or a detail must change. Focus the prompt on motion, camera, emotion/performance, and audio.

## Workflow

### Step 1 — Route the request

Decide whether the user wants:

- a video-generation prompt, or
- a character-sheet prompt for an image model

Route to the character-sheet workflow when the user wants a reusable reference sheet, turnaround, expression sheet, costume sheet, photographic identity sheet, or a consistent-character starting point for a longer image-to-video project.

If the user is asking for both, do them in this order:

1. Character sheet
2. Scene still / anchor frame
3. Video prompt

### Step 2 — If it is a video prompt, identify the model and input mode

If the user did not name a model, ask which model they are using (or offer supported options from the Model Index).

Then confirm the input mode. Start with text-to-video (t2v) or image-to-video (i2v), and use any additional keyframe or full-reference modes supported by the selected model guide.

For MiniMax H3, distinguish T2VA, I2VA, first-and-last-frame-to-video (FL2VA), last-frame-to-video (L2VA), and full-reference mode. Ask for the effective duration when a final-frame alignment or timed cuts require it.

For Seedance 2.5, distinguish text-t