MiniMax H3 Prompt Guide: 2K Video Prompts
Last updated:
Write MiniMax H3 prompts around the subject, action, setting, camera movement, lighting, and mood you want to generate. Choose text-to-video, image-to-video, first-and-last-frame, or subject-reference mode for the right level of direction.
Quick Specs
| Parameter | Specification |
|---|---|
| Generation modes | Text to Video · Image to Video · First & Last Frame · Subject Reference |
| Image inputs | jpeg, png, webp (starting frame, end frame, or subject face) |
| Text input | Natural-language scene, motion, and camera direction |
| Output resolution | Up to 2K |
| Camera language | Pan, zoom, push-in, static framing, and other prompt-level moves |
| Workflow | Write prompt → choose mode → generate → preview & download in browser |
Pick the mode that matches your control needs: text alone for speed, an image for look, two frames for a transition, or a subject photo for face consistency.
How to Choose a MiniMax H3 Generation Mode
MiniMax H3 gives you four ways to start a shot. Match the mode to how much visual control you already have—then write a focused prompt for motion, lighting, and camera.
Entry Points
- Text to Video: Use when you only have a written creative brief
- Image to Video: Use when you have a starting still to animate
- First & Last Frame: Use when you need a controlled transition between two images
- Subject Reference: Use when a face or character must stay recognizable
Examples of Prompt Direction
| Use Case | Prompt Pattern |
|---|---|
| Text concept | A rain-soaked street portrait, slow push-in, soft neon reflections, cinematic lighting |
| Animate a still | Keep the composition of the uploaded image; add gentle hair movement and a slow camera drift |
| Product reveal | Start on the closed product shot; end on the open pack; smooth dolly between the two frames |
| Subject consistency | Use the uploaded face as the subject; walk through a sunny plaza; keep facial features stable |
| Camera language | Static wide establishing shot, then a smooth zoom into the product detail |
| Mood lock | Warm golden-hour light, shallow depth of field, calm pacing, no abrupt cuts |
MiniMax H3 Prompt Formula
Core Prompt Structure
Build every brief from the same spine so the model knows what to generate and how the camera should behave:
[Subject + action] in [setting]. [Camera move], [lighting / mood]. [Optional: duration cue, aspect feel, what must stay consistent]
Copy-Paste Formula
Combine subject, scene, camera, lighting, and mode-specific guidance in one prompt:
[Subject + action] in [setting]. [Camera move], [lighting/mood]. 2K, 16:9.
Mode tip: Text only · or keep uploaded image look · or interpolate first→last frame · or lock subject face
Prompt Examples for Text to Video
1Clear Scene Direction
- Lead with who or what is on screen and what they do
- Add setting details that matter—weather, time of day, interior vs exterior
- Name one camera move (push-in, pan, static) instead of stacking five
- Close with mood and lighting so the grade stays consistent
5Product & Social Concepts
- Hero product on a clean surface with a slow orbit or push-in
- Short lifestyle beat: subject enters, interacts with product, holds for end card
- Keep scene count low—one clear action beats a crowded montage prompt
- State the share format feel (vertical social vs wide cinematic) in the brief
Prompt Examples for Image to Video
2Animate From a Still
- Preserve composition and subject identity from the uploaded still
- Add restrained motion—hair, fabric, ambient particles, soft camera drift
- Call out what must not change (logo placement, face, product angle)
- Use lighting language that matches the still so the clip does not regrade the plate
4Campaign Key Visuals
- Portrait: subtle blink and breath, slow push-in, keep background locked
- Product: soft reflection shift and a controlled reveal of packaging detail
- Illustration: gentle parallax between foreground and background layers
- End on a hold long enough for a logo or CTA if you need a social loop
Prompt Examples for Subject Reference
3Keep a Face Recognizable
- Use a well-lit, front-facing reference with few occlusions
- Describe wardrobe and location separately from facial identity
- Prefer one continuous action over many location jumps
- Re-state “keep facial features consistent” when the scene is complex
6Presenter & Character Clips
- Presenter walking and talking energy with a stable face lock
- Character enters a new environment while identity stays readable
- Soft camera follow that does not snap to a different person mid-shot
- Avoid conflicting “change the face” instructions in the same prompt
Prompt Examples for First & Last Frame
Upload an opening image and a closing image, then describe the motion that connects them. MiniMax H3 interpolates the transition while your prompt directs camera, lighting, and pacing.
First frame: closed product box on a marble table. Last frame: open box with the product centered. Slow dolly forward, warm afternoon light, soft shadow shift, clean premium look.
- · Keep both frames similar in framing so the transition stays believable
- · Name the transformation in one sentence—reveal, season shift, style change
- · Add a single camera move; avoid asking for cuts between the two frames
- · Match lighting language to both stills so the middle does not flash a new grade
Common Prompt Mistakes
- Overload the brief: Five scene changes in one short clip usually underperform one clear action with strong camera and lighting direction.
- Pick the wrong mode: If you need a controlled before/after, use First & Last Frame—not text-only. If face identity matters, use Subject Reference.
- Fight the reference image: Uploading a still then asking for a totally different composition wastes the plate. Animate what is already in frame.
- Vague camera language: "Cinematic" alone is weak. Prefer "slow push-in," "static wide," or "gentle pan left."
- Ignore consistency cues: Say what must stay locked—face, logo, product angle—especially in image and subject-reference modes.
- Use natural language: Describe the shot as you would to an editor. Clear nouns and verbs beat stacked buzzwords.
MiniMax H3 Prompt Guide FAQ
Quick answers on writing MiniMax H3 prompts—generation modes, 2K output, camera language, first-and-last-frame transitions, and subject-reference consistency.
What should a MiniMax H3 prompt include?
Start with subject and action, then setting, camera move, lighting, and mood. Add mode-specific notes: keep the uploaded image look for image-to-video, describe the transition for first-and-last-frame, or ask to keep facial features stable for subject reference. Short, concrete sentences outperform long adjective lists.
Which generation mode should I use?
Text to Video for a written concept with no assets. Image to Video when you have a starting still. First & Last Frame when you need a controlled transformation between two images. Subject Reference when a face or character must stay recognizable. Choose the smallest amount of input that still gives you the control you need.
How do I write prompts for 2K output?
You do not need special “2K keywords.” Focus on material detail, lighting, and camera intent—skin texture, product labels, soft practical lights, and a single clear move. Overcrowded prompts with many scene changes usually hurt detail more than they help.
When should I use First & Last Frame instead of Image to Video?
Use Image to Video when one still should come alive with motion. Use First & Last Frame when the story is the change between two states—product open/closed, day-to-night, style A to style B. The prompt should describe the motion path connecting those frames, not a brand-new unrelated scene.
How do I keep a character face consistent?
Use Subject Reference with a clear face photo, then write wardrobe, location, and action separately. Avoid asking the model to change identity mid-prompt. Prefer one continuous shot over many hard location jumps if likeness matters.
Can I control the camera with text alone?
Yes. MiniMax H3 responds to prompt-level camera direction such as pan, zoom, push-in, orbit, or static framing. Pick one primary move per clip so the shot stays readable.
What are the most common MiniMax H3 prompt mistakes?
Stacking too many scene changes; choosing text-only when you already have frames that should drive the result; contradicting an uploaded image; and using vague words like “epic cinematic” without naming subject, action, or camera. Fix by choosing the right mode, stating one main action, and naming one camera move.
Can I generate from text only without uploading images?
Yes. Text to Video needs no uploads and is the fastest way to explore ideas. Add an image, frame pair, or subject reference when you need tighter control over look, transition, or identity.
Ready to put these prompts to work?
Open the generator with a focused MiniMax H3 brief—or check pricing if you need more credits for longer 2K renders.