Seedance 2.5 for scenes with dialogue and characters; Kling 3.0 for short cheap single shots; Veo 3.1 for the most realistic motion from one still.

Which model for video

Picks the video model by what the shot has to do.

Use when: Before writing any video prompt.

Prompt instructions

You are my video producer. Give me a filled-in prompt that follows this guide, plus the one thing to check in the result, for a short video about [your product], aimed at [your audience]. Technique: Which model for video. Picks the video model by what the shot has to do. How it works: Seedance 2.5 for scenes with dialogue and characters; Kling 3.0 for short cheap single shots; Veo 3.1 for the most realistic motion from one still. Right for: Before writing any video prompt. What we learned from tested runs: - A tutorial's model advice is an argument, not a rule: tutorials three weeks apart named different best models for skin. - Kling 4.0 was announced on 27 Sep 2026 and was in closed beta; check availability before promising it. Be concrete: numbers, seconds and settings, no theory. Ask me one question first if something you need is missing.

Paste it into Claude, ChatGPT or Gemini.

Your productYour audience

Watch out

A tutorial's model advice is an argument, not a rule: tutorials three weeks apart named different best models for skin. Compare with one image before switching.measured: Director Chair field notes on image prompting
Kling 4.0 was announced on 27 Sep 2026 and was in closed beta; check availability before promising it.vendor: Director Chair field notes on video prompting

Notes

The choice (checked 29 Sep 2026)

The shot has to	Model
Hold characters and dialogue over 10 to 30 seconds, several shots	Seedance 2.5
Keep a real performance and change the person, the room or the costume	Seedance 2.5 video edit
Put a real person in a new setting, 3 to 10 seconds	Kling 3.0 Omni Edit
Move between a start frame and an end frame, flat animation, cheap	Kling 3.0
Several shots from one start frame, more motion, 4 to 15 seconds	Seedance 2.0
Look as real as possible in motion, 8 seconds or less, with sound	Veo 3.1

One continuous block in labelled sections: the look first, the scene, the cast, the place, the first frame and who stands where, then each shot on its own timecode, then optics, camera, physics, light, and the sound last.

Seedance 2.5Prompt

Kling 3.0Prompt

The camera first, then the action in order with its pace, then the sound in its own sentence, then what stays as in the start image, then the look sentence.

Veo 3.1Prompt

Sources

Own summary of Director Chair field notes on video prompting Checked 2026-10-13.

For agents: this card as markdown, as JSON.

Camera

Frame

Light and look

People and places

Motion

Edit

Story

Recipes

Tools