Try this in the studio
A prompt routine that saves credits
- 1
Name one subject
Say who or what is on screen. Two or three subjects give the model more ways to get it wrong.
- 2
Give one action
Say what happens, in one sentence. A clip of 4 to 15 seconds has room for one clear move, not a plot.
- 3
Add the camera and light in a few words
For example a slow push in, soft window light. Skip long lists of style words.
- 4
Say the shape
Pick 16:9, 9:16, 1:1 in the studio, and write the prompt for that frame.
- 5
Use a first frame when the look matters
A picture fixes the look of the subject, so the prompt only has to describe the motion.
- 6
Test short, then lengthen
Run 4 seconds (5 credits). When the action works, run the length you want.
- 7
Change one thing and note it
Keep a list of what you changed. You learn faster and waste less.
Three prompt shapes to start from
Starting points, not promises. Swap in your own subject and check the result.
Product on a table
A matte black coffee tin on a wooden table by a window, morning light, the camera slowly pushes in. Quiet room sound.
Person and one action
A woman in a yellow raincoat walks across a wet street at dusk, the camera follows from behind at walking pace. Rain and distant traffic.
Landscape move
A wide shot of a mountain lake at sunrise, mist drifting across the water, the camera rises slowly. Birdsong.
A cheap test plan for one new prompt
| Try | Length | Credits at 720p |
|---|---|---|
| 1 and 2: does the action work? | 4 s | 10 |
| 3: right length | 5 s | 6 |
| Total for three tries | 16 |
Sources, as checked:
- KreatorFlow MiniMax H3 page (read 17 Nov 2026): kreatorflow.ai/minimax-h3
- KreatorFlow pricing page (read 17 Nov 2026): kreatorflow.ai/pricing
Why this saves credits
Every retry is another 6 credits for a 5-second clip. A prompt with one subject and one action leaves less to go wrong, and a short test lets you find out sooner. 2K is 10 credits per 5 seconds ($0.0634/s on the API).
We have not listed H3-specific behaviour here, because MiniMax does not document it for this model. Try these habits on your own prompts and keep what works.
Prompt with a picture
MiniMax H3 takes a text prompt or one first frame. With a first frame you can describe only the motion and the sound. It saves words, and it keeps the subject the same from try to try.



