Topics
Direct answers for the H3 decisions and failure modes people search for while making a shot.
Text-to-Video or Image-to-Video?
Use Text-to-Video when the scene can be invented. Use First/Last-Frame Image-to-Video when the opening composition, subject, product, or interface must begin or end from a supplied image.
Open answerMode selectionFirst frame or first-and-last frame?
Use First/Last-Frame Image-to-Video with one image to animate away from a controlled start. Use two images when both endpoint compositions matter and H3 must create the transition.
Open answerMode selectionWhen should you use Reference Generation?
Use Reference Generation when several images, videos, or audio clips need distinct roles such as identity, wardrobe, setting, motion, or sound. It is unnecessary for a scene text alone can define.
Open answerHow to keep a character consistent in MiniMax H3
Consistency improves when one clear image owns identity, other references have narrower roles, and the prompt avoids changing wardrobe, age, lighting, and camera distance at the same time.
Open answerConsistencyHow to reduce product shape drift
Lock the product with one clean reference, request one camera move, and postpone transformations until a stable baseline preserves silhouette, labels, and material.
Open answerConsistencyHow to prevent text and logo morphing
Treat exact text as a fragile constraint. Start from a supplied frame, minimize camera and interface motion, forbid replacement text, and plan to composite critical typography in post.
Open answerHow to write camera movement in an H3 prompt
Describe camera type, direction, amplitude, speed, and timing. "Slow push toward the subject from waist-up to close-up between 2 and 6 seconds" is more executable than "cinematic camera".
Open answerCameraHow to control shot timing and cuts
Write shots in chronological order with explicit time ranges. Each shot should have one composition, one action, and at most one camera instruction.
Open answerCameraSingle take or montage for H3 video?
Choose a single take for spatial and subject continuity. Choose a montage when the story requires multiple locations, time jumps, or distinct visual beats.
Open answerHow to write dialogue for MiniMax H3
Keep dialogue short, quote the exact line, identify the speaker, and leave enough duration for the line and accompanying action. Test speech before adding a complex camera path.
Open answerAudioHow to separate ambience, effects, and music
Write diegetic ambience and action sounds with the scene, then describe non-diegetic music separately so the model does not confuse what characters hear.
Open answerAudioHow to diagnose audio sync problems
First simplify the visual action and dialogue. If sync improves, add camera movement, cuts, and secondary sounds back one at a time to identify the competing instruction.
Open answerHow to poll a MiniMax H3 API task
Create the task server-side, persist its task ID, poll at a measured interval, and stop on every documented terminal state. Do not expose the API key in client JavaScript.
Open answerDeveloperRetry strategy for failed H3 API tasks
Retry transport and rate-limit failures with bounded backoff. Do not blindly retry invalid inputs or failed generations without recording the error and changing a relevant variable.
Open answerDeveloperHow to handle H3 result URLs safely
Treat result URLs as delivery artifacts that may expire. Retrieve them after success, store only when your rights and product policy allow it, and keep generation metadata separate from the media file.
Open answerHow to estimate MiniMax H3 generation cost
Multiply the current per-second output rate by duration and attempts, then add separately priced reference inputs or upgrades. Always label the source date because pricing changes.
Open answerCostShould you iterate at 768P before 2K?
Lower-cost drafts are useful for testing story, motion, and timing. Move to 2K only after the prompt and inputs produce a candidate worth promoting.
Open answer