MiniMax H3 Video Generator turns focused inputs into polished creative results.
Prepared workflow
From brief to reviewable handoff.
Generate with MiniMax H3 in SEELE using a direct prompt handoff, H3-specific text, keyframe, and multimodal reference starters, structured output choices, and review gates.
01
Choose the H3 input path
Start with the job the shot must do. Use the text-and-audio starter when the complete scene can be described in language. Use the keyframe starter when an authorized first frame, last frame, or both must anchor composition. Use the multimodal reference starter when different images, videos, or audio files control identity, product appearance, motion, camera, performance, dialogue, music, or ambience. The on-page composer turns the selected starter and delivery choices into one structured brief and hands it directly to the MiniMax H3 template in SEELE Generation.
Text-led scene with stereo sound direction
First-to-last-frame transition
Image, video, and audio relationship brief
02
State relationships before adding detail
Name every reference by a stable label and give it one declared role. If sources disagree, say which source controls each dimension. Keep the subject, action, setting, ordered beats, camera, timing, sound, continuity, exclusions, and intended delivery frame explicit. More files do not automatically create more control; a smaller authorized pack with clear relationships is easier to interpret, review, revise, and trace than a large collection of unexplained inspiration material.
03
Generate into the intended frame
Select the deliverable and aspect ratio before continuing to Workspace so the structured request carries production context into Generation. The page supports landscape, portrait, square, or an undecided frame and distinguishes a short concept, product explainer, social cut, or editorial field note. Confirm the visible H3 model label and current input controls in Workspace, attach only rights-cleared media, then preserve the submitted prompt, model setting, references, and output for the review record.
04
Review the result before reuse
Inspect whether the result communicates the intended beat and preserves identity, geometry, camera logic, physical behavior, text, dialogue, stereo placement, timing, and continuity. Check every product fact, visible mark, voice, likeness, music source, and disclosure requirement. A successful generation is a candidate asset, not automatic evidence of accuracy, ownership, safety, policy compliance, or delivery readiness. Revise the smallest failed decision and keep rejection reasons with the output.
Capability boundary
Dated evidence
Sources behind the current facts.
Product and model details can change. These links identify the evidence checked for the claims scoped below.
Official model variants, input and output specifications, architecture, open-weight checkpoints, hosted module boundaries, recommended workflows, safety notes, and community license link.
Official distinction between base text or keyframe prompting and full-reference prompting, plus the structure of the provider-authored prompt guidance.
Reader notes
Before you hand off.
Can I generate MiniMax H3 video directly from this page?
Yes. Enter or apply an H3-specific starter, choose the deliverable and frame, then Start Creating to send the complete structured request into SEELE Generation with the MiniMax H3 template selected.
Which MiniMax H3 input modes does the page prepare?
It prepares text-and-audio, first-or-last-frame, and multimodal reference briefs. Verify the exact files, limits, controls, and entitlement shown in your current SEELE Workspace session.
Does SEELE guarantee a particular H3 result?
No. SEELE provides the integrated generation path, while every output still requires review for creative fit, identity, geometry, text, audio, factual accuracy, rights, policy, and delivery readiness.