What you will learn
- Quick Answer
- Scene formula
- Camera and motion fixes
Quick Answer
To write a controllable AI video prompt, treat it like a short production brief: subject, visible action, environment, camera movement, lighting, format, and stability constraints. Put the visible action and camera move before style words so the model knows what must happen on screen.
For Reemo video workflows, start with one scene, one subject, and one camera move. Then test the same prompt in Seedance 2.0, Kling 2.6, or Kling 3.0 and change only one variable at a time.
What You Will Learn
- How to turn a vague video idea into a structured prompt.
- Where to place subject motion, camera motion, lighting, and format.
- How to write stability constraints for product, face, label, and background.
- How to fix common prompt failures without rewriting the whole brief.
The Core Formula
Use this structure before adding decorative style words:
| Prompt part | What it controls | Example |
|---|---|---|
| Subject | What must stay recognizable | matte black travel mug |
| Action | What visibly changes | steam rises slowly |
| Environment | Where the scene happens | rainy cafe window ledge |
| Camera movement | How the viewer moves through the shot | slow push-in |
| Lighting | Mood and visibility | warm practical lights reflected on glass |
| Format | Publishing frame | 16:9 landing-page hero |
| Constraints | What must not break | keep mug shape stable, no text overlay |
Copyable Text-to-Video Template
Use this when you are starting from words only:
[Subject] in [environment], [visible action], [camera movement], [lighting], [style or realism cue], [format], keep [stable elements] consistent, avoid [failure modes].
Example:
A matte black travel mug sits on a rainy cafe window ledge, steam rises slowly, the camera pushes in, warm practical lights reflect on the glass, realistic commercial style, 16:9, keep the mug shape stable, no text overlay.
Copyable Image-to-Video Template
Use this when you already have a first frame:
Animate the uploaded image with [one subject motion], [one camera move], [lighting or reflection change], keep [product / face / label / composition] stable, realistic motion, no sudden cuts, no background replacement.
Example:
Animate the uploaded product image with a slow camera push-in, soft reflections moving across the surface, subtle background parallax, stable product shape and label position, realistic lighting, no logo changes, no sudden cuts.
How to Control Motion
Add motion in two layers. The first layer is subject motion: fabric moves gently, water droplets slide down glass, a person turns toward the window, or a product rotates slightly. The second layer is camera motion: slow push-in, dolly left, handheld drift, locked-off shot, drone pullback, or slight arc from left to right.
For first tests, choose one subject motion and one camera motion. Asking for too many scene changes in a short clip usually makes the output harder to evaluate.
How to Control Lighting
Lighting should be specific enough to guide the model. Use phrases such as soft window light, warm practical lights, neon reflections, morning sunlight through blinds, golden-hour room light, or dark studio rim light.
Avoid writing only "cinematic lighting." It can help as a style cue, but it does not tell the model where the light comes from or what should stay visible.
Troubleshooting Table
| Problem | Rewrite |
|---|---|
| The clip looks stylish but nothing happens | Add one visible subject action near the start |
| The camera movement feels random | Replace vague motion with slow push-in, locked-off shot, or handheld drift |
| The product shape changes | Add keep product shape, label position, and composition stable |
| The scene changes too much | Add one scene only, no sudden cuts, no background replacement |
| The format crops the subject | Add 9:16, 16:9, 1:1, or 4:5 and name the main subject position |
FAQ
What is the best structure for an AI video prompt? Use subject, action, environment, camera movement, lighting, format, and constraints. This gives the model both creative direction and stability instructions.
Should I write style words first? No. Put the subject, action, and camera move first. Style words such as cinematic, realistic, premium, or UGC are useful only after the model knows what should move.
How many camera moves should I ask for? Start with one. A slow push-in, locked-off shot, handheld drift, or simple dolly move is easier to evaluate than several camera moves in one short clip.
How do I keep a product or face stable in image-to-video? Name the stable elements directly: keep the product shape, label position, face identity, composition, and background consistent. Also say no sudden cuts or background replacement.
Which Reemo video model should I test first? For structured movement and stability, start with Kling 2.6 or Kling 3.0. For cinematic exploration, test Seedance 2.0. If the project matters, run the same prompt in more than one model.
Methodology and Update Notes
This lesson is based on Reemo editorial prompt testing for text-to-video and image-to-video workflows. It focuses on prompt structure, stability language, and practical review criteria rather than ranking third-party demos.
Last reviewed: June 25, 2026. Updated to the Academy GEO template with Quick Answer, formula tables, copyable templates, troubleshooting, FAQ, and methodology notes.