H3 Max Prompt Guide

How to prompt H3 Max in five steps

Start with the shot, not a pile of keywords. H3 Max has to decide what appears, what changes over time, where the camera goes, and what the scene sounds like. A prompt becomes easier to follow when each sentence answers one of those production questions.

1. Choose a generation mode. Use Text to Video when the scene starts from words, or Image to Video when an uploaded frame should define the composition.
2. Define one shot. Name the subject, setting, and one action that can finish within 5 to 15 seconds.
3. Direct the camera. Choose one primary framing and camera movement.
4. Add the visual and audio treatment. Specify lighting, palette, ambience, effects, dialogue, or music only when relevant.
5. Set and test. Choose duration, resolution, aspect ratio where available, and prompt expansion before generating.

Fast template: [subject and place]. [one visible action]. [shot size and camera movement]. [lighting, color, texture]. Audio: [ambience, effects, dialogue, music choice].

Then match the idea to the available controls: 5–15 seconds, 480P or 768P, and—on Text to Video—one of six aspect ratios. The prompt expansion setting can preserve your wording or elaborate it. Those settings are part of the creative direction, not an afterthought.

The five-part H3 Max prompt formula

The formula is a planning tool, not rigid syntax. Write it as natural language. The goal is to make every important instruction concrete enough to become a frame, a movement, or a sound.

1. Subject and scene

Name the subject with one or two distinguishing traits, then locate it. “A cyclist” leaves clothing, age, bicycle, weather, and surroundings open. “A courier in a yellow rain shell waiting beside a cargo bike under a concrete overpass” establishes a coherent starting frame.

Do not overload the first sentence with biography. H3 Max needs what the camera can see. Replace abstract character notes with visible evidence: tense shoulders, paint-stained hands, a dented helmet, or a careful grip.

2. Visible action

Use active verbs and a small number of beats. Short clips reward causality: the subject does something, the environment reacts, and the shot reaches a stopping point. “The courier checks the sky, zips the jacket, and pushes away as rain begins” is a sequence the viewer can read.

3. Camera

Specify framing before movement: close-up, medium shot, wide shot, overhead, low angle, or point of view. Then add one main movement such as a pan, tilt, dolly, orbit, tracking move, push-in, or pull-back. If you want no movement, say “locked camera” rather than leaving the choice open.

4. Visual treatment

Describe light and color through physical sources: late-afternoon side light, sodium street lamps, a flickering monitor, a hard spotlight, or overcast daylight. Material and atmosphere words—wet asphalt, dusty air, fine film grain, crisp product reflections—usually communicate more than a long list of genre adjectives.

5. Audio

Place audio at the end so it reads as its own direction. List foreground effects, background ambience, spoken words, and music choice. The site generates audio with the video, but generated dialogue and synchronization can vary, so concise lines and visible speakers are easier to evaluate.

H3 Max text-to-video prompt examples

Text mode asks the prompt to establish the whole opening image. Put the strongest compositional information early: who or what we see, the location, time of day, and initial shot size. Style should support that scene rather than replace it.

01

Quiet product reveal

Text to Video · 16:9 · 8 seconds · 768P

Why it works: One subject, one controlled movement, and one camera instruction keep the shot legible. The sound cues belong to visible objects, so they support the scene instead of competing with it.

Recreate

02

Street-food documentary

Text to Video · 16:9 · 10 seconds · 768P

Why it works: The prompt gives the camera a three-beat route without inventing multiple locations. The 16:9 frame keeps the cook, griddle, customer, and rainy market context readable together.

Recreate

03

Single-action character shot

Text to Video · 16:9 · 6 seconds · 768P

Why it works: The action has a clear endpoint. “Stops just before touching” gives the clip a readable finish rather than asking the model to improvise an ending.

Recreate

04

Architectural atmosphere

Text to Video · 16:9 · 12 seconds · 768P

Why it works: Environmental motion makes a mostly static scene feel alive. The wide 16:9 composition is established before the centered camera move.

Recreate

Turn a concept into a shot

Suppose the concept is “a futuristic train station.” That names a setting, not a video. Choose a subject, change, and viewpoint: “A lone commuter steps from a silent magnetic train as hundreds of paper tickets lift from the floor in its wake. Start in a low wide shot beside the platform edge, then track backward in front of the commuter. Cold white ceiling panels, cobalt signage, polished black floor. Audio: soft train brakes, rushing paper, distant announcements.”

The rewritten version gives the station scale through a person, creates movement through the tickets, and keeps the camera on one path. It also avoids asking for an entire story in ten seconds.

Write timing as beats, not edits

For a single continuous shot, time can be expressed with plain transitions: “At first,” “as the door opens,” “then,” and “finally.” Use timestamps only when exact staging truly matters, and keep enough time for each beat to be visible. If a six-second prompt contains an establishing shot, three actions, two camera moves, and an ending reveal, compression—not missing adjectives—is the problem.

Apply the Formula to Image-to-Video

In image mode, the upload already provides identity, wardrobe, composition, color, and spatial relationships. The best prompt usually describes the change from that first frame. Tell H3 Max what moves, what stays stable, how the camera behaves, and how the shot ends.

This guide teaches the method and keeps two focused image examples beside the controls they demonstrate. For a larger swipe file organized around motion patterns and failure fixes, use the 15 H3 Max image-to-video prompts.

05

Animate a portrait

Image to Video · 16:9 · 5 seconds · 768P

Video result

Why it works: The uploaded image already defines the person and composition. The prompt spends its words on restrained motion, preservation, and sound instead of redescribing the portrait.

Recreate

06

Animate a landscape

Image to Video · 16:9 · 8 seconds · 768P

Video result

Why it works: The directions separate subject motion, environmental motion, and camera motion. Preservation language protects the geometry already present in the source image.

Recreate

Use the image as a contract

If a portrait already shows a blue coat, a brick wall, and window light, repeating all three consumes attention without adding motion. Use preservation instructions only for features that matter: “preserve facial features and coat design,” “keep the logo readable,” or “maintain the original camera angle.” Then spend the rest of the prompt on breathing, gaze, fabric, weather, background activity, and camera drift.

Optional end frames need a transition

When the interface offers an end image, do not merely describe both pictures. Describe how the first state becomes the last: the direction of travel, the object that changes, and the camera path connecting them. Large identity or geometry changes are harder than a continuous action between compatible compositions.

The output follows the uploaded image’s shape in Image to Video. Crop the source to the intended orientation before upload; the text-mode aspect-ratio buttons do not override an image’s framing.

Reference-to-Video prompting

Reference mode starts with uploaded images, videos, or audio rather than a blank prompt. Add the materials first, type @ in the prompt editor to select a named reference such as Image 1 or Video 1, then describe the role that reference should play, the action, camera behavior, and ending state. Mention only references that affect the shot so the instruction stays direct.

Camera and motion vocabulary that stays clear

Camera words work best when they describe a physical relationship. “Cinematic” is a judgment; “a slow shoulder-height tracking shot two meters behind the runner” is an instruction. Pick a move that reveals information or follows action.

Separate subject motion from camera motion

“The dancer spins as the camera orbits counterclockwise” contains two rotations that may fight for attention. If both are essential, slow one of them and state their relationship. Otherwise let the subject move against a locked camera or track in a simple direction.

Control motion intensity

Modifiers such as barely, gentle, steady, brisk, sudden, and accelerating change the energy of a shot. Use them consistently. “A calm scene with a rapid whip pan and violent handheld shake” sends mixed signals unless the contrast is the point of the story.

How to write audio into the prompt

To make audio direction easier to evaluate, start with the sound caused by the visible action, then add one layer of ambience. Music and dialogue are optional. Silence can also be directed: “no music,” “no speech,” or “only room tone and the fluorescent buzz.”

• Effects: footsteps on tile, a latch click, fabric movement, engine idle, glass set on wood.
• Ambience: distant traffic, coastal wind, a crowded room, insects at dusk, a quiet studio.
• Dialogue: identify the speaker and keep the line short. Avoid several people speaking over one another in a brief clip.
• Music: specify presence, absence, and broad character. Do not rely on a named copyrighted song as the whole direction.

Audio pattern: Audio: [synchronized foreground effect], [background ambience], [short dialogue if needed], [music or no music].

When sound is not important, a restrained instruction is better than filling the prompt with audio adjectives. The model still has to balance the soundtrack with visual generation.

Settings that change how a prompt behaves

Duration: 5 to 15 seconds

Duration determines how much action has room to unfold. Use five or six seconds for one gesture, a product movement, or a compact reveal. Eight to ten seconds can support two or three connected beats. Twelve to fifteen seconds gives a tracking shot or environmental change more breathing room, but it also creates more time for composition to drift.

Resolution: 480P or 768P

If lower-resolution drafts fit your review needs, test the core idea at 480P and move to 768P when higher output detail is required. On this site, 480P uses one credit per second and 768P uses two. Resolution does not repair an overloaded prompt; it changes output detail and credit cost.

Aspect ratio in Text to Video

Choose 16:9 for general landscape video, 9:16 for vertical social framing, 1:1 for square placements, 4:3 or 3:4 for more compact compositions, and 21:9 for an ultrawide frame. Mention compositional intent that suits the ratio—full-body vertical action, centered square product, or panoramic environment—rather than expecting a late crop to preserve every element.

Prompt expansion

Disabled keeps authorship closest to your wording and is useful for tightly directed shots. Balanced is a practical default when your structure is clear but the scene could use modest interpretation. Quality can elaborate a short seed, but added detail may compete with exact composition or identity. If a result wanders, simplify the prompt and reduce expansion before adding more constraints.

For the exact pixel dimensions, frame rate, aspect-ratio behavior, and credit math, use the H3 Max resolution guide. Keeping those specifications on one page prevents this prompt tutorial from becoming a duplicate model overview.

Why H3 Max prompts fail—and what to change

Revise one variable at a time

If you change the action, camera, duration, ratio, and expansion together, the next result cannot tell you which change helped. Keep the seed idea stable and revise the highest-priority failure first. Prompting becomes a repeatable craft when each generation tests a hypothesis.

A repeatable H3 Max prompting workflow

1. Write the outcome in one sentence. What should the viewer understand or feel when the clip ends?
2. Choose the mode. Use Text to Video for invention; use Image to Video when a source frame must anchor the scene.
3. Draft with the five-part formula. Subject, action, camera, look, audio. Cut repeated adjectives, but keep timing, preservation, and exclusion constraints that make the intended result testable.
4. Match action to time. Count the visible beats and give each enough seconds to read.
5. Choose a review setting. If 480P is sufficient for judging motion and composition, a 5-second draft uses 5 credits. Select 768P when the review or final placement needs its higher output detail.
6. Diagnose the biggest miss. Identity, action, camera, environment, or sound—not “quality” as a vague category.
7. Revise one layer. Shorten an action, lock the camera, protect a feature, or adjust expansion. Keep what already works.
8. Generate the delivery framing. Move to the required length, resolution, and delivery aspect ratio after the direction is stable.

Save successful prompts with their mode, duration, resolution, ratio, and expansion setting. The wording alone does not fully describe a result. A useful prompt library records the production choices that made the prompt work.