MiniMax H3 Reference to Video: A Character Consistency Guide

Give each reference a clear job

MiniMax H3 reference to video starts with a useful question: what should each input control? A character picture explains appearance; a movement clip shows a performance; an audio reference supplies sound-related direction. A first-frame image has a different job: it establishes the opening composition. MiniMax documents these as separate generation modes in its official video guide.

For character consistency, begin with a readable identity and one clear scene. Adding inputs is useful when they resolve a specific uncertainty. A larger collection also gives you more details to reconcile, from conflicting outfits to different lighting and camera angles.

Choose the workflow before choosing references

ReelBoop’s Photo to Video tool includes MiniMax H3 Max Turbo in its model selector. Its current workflow takes one photo and a motion prompt. Character Swap currently uses PixVerse, while Body Swap uses Kling v2.6 Standard motion control. For the upload steps and credit rules, use the video character replacement tutorial.

What the official H3 reference interface supports

MiniMax’s V2 API lists four model IDs: MiniMax-H3, MiniMax-H3-Highspeed, MiniMax-H3-Lite and MiniMax-H3-Max. Reference inputs use explicit image, video and audio roles. First/last-frame inputs belong to the image-to-video route, and the API keeps those roles separate from reference roles. See the V2 request specification.

For reference generation, the official guide currently lists up to nine images, three video clips and three audio clips, with a combined limit of twelve files. Those are interface ceilings, not a recommended recipe for one character. Read the selected model’s current requirements before preparing a batch. MiniMax input requirements

The provider matters too. ReelBoop’s H3 choice maps to fal’s minimax/h3-max-turbo/image-to-video, a separately documented variant. The fal endpoint documentation describes a first-frame image input. ReelBoop exposes its single-photo workflow; its form does not offer multiple H3 character-reference slots, a reference-video upload or an audio upload.

Build a character reference you can judge

Pick one anchor photograph before writing a scene. At the size you expect to publish, you should be able to recognize the face, hairstyle and overall silhouette. For a standing performance, show the full body. For a quiet close-up, prioritize facial detail and even light.

Check four details in that anchor:

• Face: visible eyes and facial structure, with enough detail beyond a heavily smoothed filter.
• Hair: a readable hairline, length and shape.
• Clothing: the outfit you want to carry through the sequence.
• Framing: enough room for the action without immediately cutting off hands or feet.

Keep the original file available for every shot. A new crop, a stronger beauty filter or a different outfit introduces another variable when you compare results.

Can several pictures show the same character?

In a compatible H3 reference interface, several images can fit within the documented reference-image allowance. Choose each for a reason: perhaps a front view makes the face clear while a side view explains the hair silhouette. Look for complementary information and resolve conflicting views before adding more.

Lay the pictures side by side first. Do they show the same age, face shape, hairstyle and outfit? If one image contradicts the others, decide which look you actually want. In ReelBoop’s Photo to Video tool, choose the single photograph that best explains your planned shot.

Write one performance you can recognize

A practical motion prompt describes the subject’s action, the camera and the pace. Begin with one action that can be checked during playback. For a first-frame photo workflow, try a direction such as:

The person gives a small smile, turns slightly toward the window and pauses. The camera remains at eye level with a gentle push-in. Keep the hairstyle, white shirt and soft daylight coherent throughout the shot.

Adapt the clothing and setting to your actual photo. Check the expression, turn and camera move separately during playback. If you need an exact supplied performance, choose a motion-video workflow instead of describing complex choreography in a paragraph.

Plan consistency across three shots

Write a small shot list before generating: a medium shot with a wave, a closer shot with a glance, then a wider shot with a small turn. Reuse the same anchor photo, outfit description and lighting direction. Give each shot one distinct action.

This makes the review concrete. Compare the face at similar angles, then look at hair, neckline, clothing color and proportions. Judge motion separately: a recognizable person can still move awkwardly, and a smooth movement can still change the character’s appearance.

Decide where the soundtrack belongs

An audio reference is an input to a supported H3 reference-generation interface. Uploading a known soundtrack is a separate product choice, and ReelBoop’s Photo to Video tool currently has no audio-upload control. If your edit needs a specific track, plan to add it in your editor and check the final timing there.

For a preset dance, read that template’s sound information. A general model’s audio capabilities do not tell you what an individual template delivers.

Common questions

Does a first frame lock the character for every later shot?

It provides the starting image. Review identity through the whole clip and compare separate shots before assembling a series.

Can I use H3 to replace a person on ReelBoop?

Use Character Swap for that task. The current tool has its own model, source-video limits and credit quote. The H3 photo option is for generating motion from a still image.

Where should I start with one good photo?

Open Photo to Video, choose the model, describe one small action and review the displayed credit cost. A clear first shot gives you a useful baseline for the next creative decision.