AI video prompting becomes much easier when you stop treating the prompt as a visual description and start treating it like a short production brief.
I’ve been experimenting with MiniMax H3, and a structure that works well is:
Subject + Action + Camera + Environment + Lighting + Audio + Constraints
For example:
Create a cinematic product video featuring a silver wireless speaker on a dark wooden desk. Start with an extreme close-up of the speaker texture, then slowly dolly backward to reveal the full product. Warm afternoon sunlight enters from the left, creating soft shadows. Add subtle room ambience and a quiet electronic startup sound. Keep the product shape, logo, and proportions consistent throughout the shot. Avoid extra objects, distorted text, sudden camera movements, or unnecessary transitions.
The important part is separating each instruction by purpose.
1. Define the subject clearly
Instead of:
A cool product on a desk.
Try:
A compact matte-black wireless speaker with a silver control ring, centered on a walnut desk.
2. Describe motion separately
Tell the model exactly what should move:
The camera performs a slow 20-degree orbit while the product remains stationary.
This usually gives more predictable results than simply asking for a “cinematic shot.”
3. Give audio its own direction
When generating video with audio, describe sound just as deliberately as the visuals:
Audio: quiet studio ambience, a subtle mechanical click when the device powers on, no music and no voiceover.
4. Add consistency constraints
For reference-based videos, explicitly tell the model what must stay unchanged:
Preserve the character's face, hairstyle, outfit, proportions, and color palette from the reference image.
This becomes especially useful when combining image, video, or audio references.
Reusable MiniMax H3 Prompt Template
Subject: [main character, object, or product]
Action: [exact movement or event]
Camera: [shot size + camera movement]
Environment: [location + background]
Lighting: [light direction + mood]
Style: [cinematic, commercial, anime, documentary, etc.]
Audio: [dialogue, ambience, music, SFX]
References: [what each uploaded reference controls]
Constraints: [what must remain consistent + what to avoid]
The biggest improvement usually comes from being specific about motion, reference roles, and audio, rather than adding more style adjectives.
For anyone experimenting with MiniMax H3 prompts, I’ve also been using MiniMax H3 Video to test text-to-video, image-to-video, and reference-based workflows:
The example gallery is useful because the prompts are shown alongside the generated video, which makes it easier to study how different instructions affect the result.
Would be interested to see what prompt structures other people are using for character consistency or reference-to-video workflows.
Top comments (0)