MiniMax H3 Max Ref usually refers to the reference-to-video workflow in fal's H3 Max family. If you want a recurring character, a recognizable prop or a shared visual style across scenes, start by deciding what each reference should control. A pile of attractive images is less useful than a small reference set with a clear purpose.
What MiniMax H3 Max Ref means
fal's H3 Max reference-to-video documentation identifies H3 Max as its post-trained variant of MiniMax H3. The schema lets a prompt refer to assets by type and order, such as Image 1, Video 1 and Audio 1. It documents a combined maximum of 12 reference files across those types. Treat those as provider-specific capabilities; a different H3 interface may expose different controls.
The task is to generate a scene guided by references. If you already have a finished source clip and want to replace a person in that clip, use the MiniMax H3 Recast guide to plan the cast edit. Reference-led generation and source-video recasting solve different creative problems.
Make a MiniMax H3 Max Ref asset-role sheet
Before selecting files, write a one-sentence shot brief: who is present, what happens, where the camera is, and what the audience should hear. Next assign one job to each reference. A portrait may define the protagonist's appearance; a room photo may define the setting; a short movement reference may communicate the rhythm of an action.
Name files descriptively and make a matching asset list in your notes. For example, label a portrait as "protagonist-front" and a room image as "workshop-lighting." The names are for your organization; the prompt should follow the selected interface's actual reference labels. This separation stops accidental reordering from changing the intended meaning of a prompt.
Remove redundant files before adding more. Two portraits with substantially different hair, age, accessories or styling can create an unclear identity brief. Decide which characteristics are essential and select an internally consistent set. Use photographs and recordings you are allowed to use, especially when an identifiable person or voice is involved.
Write a scene prompt with reference responsibilities
Our original planning example separates identity, setting, camera and sound. It is a writing exercise, not a claim that this exact prompt has been generated or benchmarked:
Image 1 defines the protagonist's face and hair.
Image 2 defines the workshop's warm side lighting and wooden workbench.
The protagonist sets a small ceramic cup on the bench, then looks toward the door.
A steady medium shot follows the hand movement without changing the viewpoint.
Sound: a soft ceramic tap, quiet room ambience and a faint door creak.
Prefer one coherent action over several unrelated events. If the shot needs an entrance, an object exchange and a dramatic camera move, decide which event carries the story. Split the remaining events into another shot brief. A short, clear scene is easier to review and revise than a prompt where every sentence introduces another visual demand.
Keep identity instructions distinct from action instructions. "Use the face from Image 1" tells the model what should persist. "Turns toward the door" tells it what should change. This distinction also makes your review notes more specific: you can identify a resemblance problem without rewriting an otherwise successful action.
Plan continuity across a three-shot sequence
Create a compact character sheet with the attributes that matter to your scene: hairstyle, clothing, dominant accessory and any prop that must remain recognizable. Carry the same wording and reference order into every shot. Describe the changing action and camera position separately rather than rewriting the character from scratch.
For a workshop sequence, one shot could establish the protagonist at the bench, a second could focus on placing the cup, and a third could show the reaction to a sound at the door. Keep the room's light direction and the cup's position in the notes. Continuity includes objects and geography, not just the face.
After generating a candidate, compare the opening and closing frames with the character sheet. Check the eye area, hair silhouette, clothing details, object shape and contact between hands and props. Watch again at normal speed for temporal drift. Do not accept a clip solely because one screenshot resembles the reference.
MiniMax H3 Max Ref FAQ
What does Ref stand for?
In this search context, Ref is shorthand for reference-to-video. Use the full endpoint or interface name when comparing features so it is clear which provider and workflow you mean.
Is H3 Max a different name for every MiniMax H3 model?
No. fal describes H3 Max as its post-trained variant. Do not assume that every interface using the MiniMax H3 name has the same reference inputs or settings.
How many references should I start with?
Begin with the fewest files needed to define the subject and scene. Add another only when it has a distinct role, then check the selected provider limits.
Can a reference portrait replace a person in a finished clip?
That is a source-video recasting task. A reference-led scene brief is intended for a newly generated scene; choose the workflow that matches your starting material.
How do I compare candidate clips?
Use the same character sheet and review moments for each clip. Assess identity, prop contact, motion continuity and sound independently rather than choosing from a single attractive frame.
Build your next shot
Use the H3 prompt guide to turn a rough idea into a scene with action and sound. Explore H3 video examples, then create an H3 video using the inputs available in your selected tool. Save the reference-role sheet with the result so the next scene starts from the same creative brief.
Nolan Reeve