# MiniMax H3 Prompt Engineering: Camera Motion, Timing, and Native Audio > Source: > Published: 2026-09-19 03:09:47+00:00 Generating a visually attractive AI video is easy to describe, but much harder to control. With MiniMax H3, the difference between an average result and a usable shot often comes down to how the prompt describes four things: Instead of treating a prompt as one long visual description, it is more useful to think of it as a small production specification. This guide explains a practical structure for writing MiniMax H3 prompts that are easier to control and reuse. A useful MiniMax H3 prompt can be structured like this: ``` text Subject + Action + Environment + Camera movement + Lighting + Visual style + Timing + Audio For example: A woman wearing futuristic silver sunglasses stands on a rooftop at sunset. She slowly turns toward the camera while the wind moves her hair. The camera performs a slow cinematic dolly-in from a medium shot to a close-up. Warm sunset light reflects from the glasses, with soft rim lighting around the subject. Luxury fashion campaign aesthetic, realistic skin texture, shallow depth of field. The movement begins slowly, accelerates slightly in the middle, and ends with the subject holding eye contact with the camera. Ambient city sounds, light wind, and subtle cinematic background music. The important idea is that every sentence has a job. This makes the prompt much easier to debug than a single paragraph full of adjectives. 2. Describe Motion as a Timeline One of the most common prompting mistakes is describing only what the final frame should look like. Video models need information about what happens between frames. Instead of: A sports car driving through Tokyo at night. Try: A black sports car accelerates through a neon-lit Tokyo street. 0–2 seconds: The car enters the frame from the left while the camera tracks alongside it. 2–4 seconds: The camera moves slightly lower and closer to the front wheel as reflections from neon signs move across the bodywork. 4–6 seconds: The camera pulls back into a wider tracking shot while the car turns through an intersection. Even when exact timestamps are not strictly required, thinking in this format forces you to describe temporal progression. This is particularly useful for: product reveals fashion shots transformation sequences vehicle videos character movement looping clips 3. Separate Subject Motion From Camera Motion Another frequent problem is asking the subject and camera to perform too many movements simultaneously. Consider this prompt: The woman walks forward while turning around as the camera circles her and zooms in quickly. There are multiple competing motion instructions. A more controlled version would be: The woman walks slowly toward the camera. The camera tracks backward at the same speed, maintaining a medium shot. During the final two seconds, the camera performs a gentle 15-degree arc toward her right side. This gives the model a clearer motion hierarchy. I normally think about motion in two layers: Subject motion: walk / turn / reach / look / pick up / rotate / jump Camera motion: dolly-in / dolly-out / pan / tilt / orbit / tracking / crane / handheld Choose one dominant movement from each layer before adding anything more complicated. 4. Use Camera Language Precisely Generic phrases such as: cinematic camera movement leave a lot of interpretation to the model. More specific instructions are usually easier to control: slow dolly-in left-to-right tracking shot 90-degree clockwise orbit low-angle forward tracking shot static close-up with subtle handheld movement slow crane upward revealing the skyline It also helps to define framing: extreme close-up close-up medium shot full-body shot wide shot aerial shot For example: Begin with a wide establishing shot. Slowly dolly forward toward the subject. Transition naturally into a medium shot without cutting. That is much more specific than simply saying "cinematic zoom." 5. Treat Product Videos Differently For commercial product videos, excessive subject movement is often unnecessary. The object should usually remain easy to recognize. Example: A premium black perfume bottle sits on a reflective stone pedestal in a dark studio. The bottle remains stationary. The camera performs a slow 40-degree clockwise orbit while gradually moving closer. A narrow beam of warm light travels across the glass surface, revealing the embossed logo. Fine atmospheric particles are visible in the background. Luxury fragrance advertisement, realistic reflections, dark cinematic lighting, premium commercial photography. Subtle ambient sound and a soft glass shimmer at the end. Most of the perceived motion can come from: camera movement lighting reflections environmental particles This often preserves product consistency better than asking the product itself to perform complicated transformations. 6. Write Audio as Part of the Scene If the workflow supports native audio, audio should not be treated as an afterthought. For example: Audio: quiet café ambience, soft conversation in the background, ceramic cup placed gently on a wooden table, subtle acoustic music. For an action scene: Audio: engine acceleration, wet tire sounds, rain hitting the windshield, distant city ambience, no dialogue. When speech is not required, explicitly writing: No dialogue. can remove ambiguity from the prompt. 7. Use Constraints Sparingly Negative instructions are useful, but adding too many can make a prompt harder to interpret. Instead of: no blur, no distortion, no flickering, no morphing, no camera shake, no duplicate objects, no incorrect hands, no changing clothes... focus on the failure modes that matter for the specific shot. For example: Keep the sunglasses unchanged throughout the shot. Maintain consistent facial identity. No scene cuts. These constraints directly protect the important elements of the sequence. 8. A Reusable MiniMax H3 Prompt Template Here is a compact structure that works well for experimentation: SUBJECT: [Who or what is visible?] ACTION: [What changes during the shot?] ENVIRONMENT: [Where does the scene happen?] CAMERA: [Framing + camera movement + direction + speed] LIGHTING: [Main lighting characteristics] STYLE: [Commercial / cinematic / realistic / documentary / fashion / etc.] TIMELINE: [Beginning → middle → ending] AUDIO: [Ambient sound + effects + music + dialogue] CONSTRAINTS: [Only the most important consistency requirements] You can expand or remove sections depending on the shot. 9. Example: Simple Prompt vs Structured Prompt Simple A futuristic sneaker commercial with dramatic lighting. Structured A white futuristic sneaker rests on a matte black pedestal inside a dark studio. The sneaker remains stationary. The camera begins with a low-angle close-up of the sole and slowly performs a clockwise orbit while pulling back into a three-quarter product shot. A narrow blue-white studio light sweeps from the heel toward the toe, revealing the texture of the material. Fine atmospheric haze creates subtle light beams behind the product. Premium sportswear advertising aesthetic, realistic product photography, controlled reflections, high contrast. The shot begins almost completely dark, reveals the sneaker progressively, and finishes with the entire product clearly visible. Audio: low atmospheric bass, subtle mechanical texture, soft impact sound at the final reveal. Keep the shoe shape, color, logo placement, and proportions consistent throughout the shot. The second version gives the model significantly more information about how the video should evolve. 10. Build a Prompt Library Instead of Starting From Zero Once you find structures that work, save them by use case instead of writing every prompt from scratch. Useful categories include: product advertising fashion food automotive cinematic character shots social media ads image-to-video animation camera motion tests More MiniMax H3 examples, workflows, and video generation tools: https://minimax3.org/ Final Thought The most useful shift in AI video prompting is to stop describing only an image. Describe a shot. A shot contains: subject + movement + camera + time + sound Once those elements are explicit, prompts become easier to modify, compare, and reuse. For more MiniMax H3 resources: https://minimax3.org/ ```