Basic Rules

1. Anchor the character and environment first

Briefly confirm the main character, appearance, position, and environment before describing motion. This helps maintain better consistency.

You don't need to describe the whole image again — <Picture 1> is already the starting frame.

2. Use shots or timestamps for better control

Use:

[Shot 1] → [Shot 2]

to separate different camera shots.

For more precise motion timing, you can also describe actions with time ranges such as:

[0s~2s] → [2s~4s]

Keep it simple unless you really need precise control.

3. Don't ignore the audio

H3 can generate both video and sound.

Describe the environmental sound, action sounds, dialogue, and music when needed. Clear audio instructions can prevent random or unsuitable sound generation.

Recommended Full Format

For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.

integrated_multimodal_description:
[Shot 1] [visual style], the [subject] shown in <Picture 1> preserves [important appearance / clothing / position / environment]. [Initial action]. [Main action develops]. [Secondary physical or environmental motion]. The camera [camera movement + speed/amplitude if needed]. [Final pose, expression, or result].

overall_soundscape:
[environmental ambience]. [sounds caused by visible actions].

non_diegetic_music:
[music description or None].

IF You Are Lazy ...

Please use (749) Enjoy your imagination (Cash version) | Tensor.Art | Tensor.Art

Prompt the scene right and it'll do it for you :) .

Thanks for reading! I’m still testing and learning more about H3, so I’ll keep updating this guide whenever I find something useful. Hope it saves you some trial and error — happy generating!

Reference : docs/VIDEO_PROMPT_WRITING_GUIDE_base_ · MiniMaxAI/MiniMax-H3 at main