How to Write Prompts for the Five Modes

Text-to-video, creative long video, manuscript video, poetry video and digital anchor — how to prompt each one

Before you write, decide which mode you want: text-to-video (a single description, optionally with a reference image to lock the subject), creative long video (a storyline plus a scene list), manuscript video (a whole article auto-split), poetry video (a poem split line by line) or the digital anchor (a reference portrait plus a voiceover script; narration and subtitles are generated automatically — see the "Narration & Subtitles" guide). Each mode asks for a different emphasis in the prompt. All modes share the same limits: roughly 16 requests per minute, up to 720p, and a maximum of 20 seconds per clip.

Five modes, five approaches

Let the mode decide the focus; then add the general tips below.

1

Text-to-video (single description)

Describe the shot in one paragraph: subject, action, scene, light. Write as if narrating a line of voiceover — every phrase you type should be drawable. To lock the subject, attach a reference image (image-to-video).

Example: simple text-to-video
2

Creative long video (multi-shot)

Write a one-line story idea, then a scene-by-scene shot list: time, who, what, mood. The model renders each scene and stitches them into a long clip, great for stories and explainers.

Example: creative long video
3

Manuscript long video

Paste a whole article or script: AI splits it by reading duration, generates visuals per segment, then unifies narration and subtitles before stitching. Ideal for explainers and course content.

Example: manuscript long video
4

Poetry video

Paste a poem, AI splits it line by line: the original verse becomes narration and subtitles, visuals are generated per line, then stitched into an atmospheric short. Great for classical poetry and mood animation.

Example: poetry video
5

Digital anchor

Reference portrait + a voiceover script. The image decides the look, the script decides the tone; narration and subtitles come automatically. Ideal for talking-head reviews, tutorials and intros.

Example: digital anchor

General prompting tips

  • Be concrete

    Turn a vague idea into something visible. Instead of "a beautiful girl", say "a girl in a red dress running on tiptoe across a rainy platform". If it can't be drawn, it won't be rendered well.

  • One shot per sentence

    Use periods and semicolons to break your prompt into shots and beats, one action per sentence. Don't jam a whole story into a single action phrase.

  • Describe light, camera and depth

    Say the light source, contrast, depth of field and camera angle, e.g., warm backlight, shallow depth of field, dutch angle. If light and angles work, the frame looks instantly better.

  • Set the style in the first line

    Open with one sentence that fixes the whole look, for example cyberpunk neon or ink-painting. Once the style is locked, later shots keep the vibe.

Common pitfalls

  • Too many shots in one prompt

    Explosion + car chase + romance packed into a single sentence makes the model grab one randomly. Write one shot at a time; use creative mode for the rest.

  • Adjectives instead of pictures

    "Beautiful", "epic", "cool" are reviews, not directions. Translate them into light, framing, action and composition so the model can shoot them.

  • Ignoring subject consistency

    If a character keeps changing look across shots, you likely forgot to describe the main subject. Give a reference image or repeat key identity details to anchor them.

Try one line in the Demo

Write one prompt following the tips, paste it into the online demo and it will be ready to render. Free.

Open demo →

Prompt-Methodology References

The prompt-writing approaches here are grounded in the following model technical reports and prompt-engineering guides.

  1. VideoPoet: A Large Language Model for Zero-Shot Video GenerationGoogle Research · arXiv:2312.14125 (2023)

    The VideoPoet paper demonstrates the sensitivity of LLM-driven video generation to prompt structure, providing academic grounding for the prompt tips on this page.

  2. Sora: Video Generation Model Technical ReportOpenAI · OpenAI Technical Report (2024)

    OpenAI's Sora technical report details how prompt engineering affects video quality, including best practices for camera language and physical descriptions.

  3. OpenAI Prompt Engineering Guide — 提示词工程最佳实践OpenAI · OpenAI Docs (2025)

    OpenAI's official prompt-engineering guide offers general methods — be specific, step-by-step, constrained — consistent with this page's tips.

Why you can trust this

Written by a practicing developer and cross-checked against authoritative primary sources such as arXiv papers, vendor technical reports, and the Stanford AI Index.

S

SandGrid@lcy362

Author of Agnes Video Generator · Full-stack Developer

Independent developer and author of Agnes Video Generator, an open-source (MIT) AI video generation tool built on Agnes AI's free video models. Has helped 1,000+ creators produce AI videos at zero cost. Focused on making video generation models accessible and production-ready; content is based on hands-on practice and primary research.

View on GitHub

Last updated2026-08-21

Ready to Start Creating?

"Making world-class AI belong to everyone." — Bruce Yang. It's completely free, no credit card, and you won't need a high-end GPU. Your first AI video starts at zero cost. Want to use Agnes AI's free video models? This is the easiest way in.

Clone the GitHub repo and launch in 2 minutes