MiniMax H3：帶有立體聲音訊的 768p 和 2K 文字轉視頻 |運行舒適 | Models and API | RunComfy

Why Choose MiniMax H3#

MiniMax H3 is MiniMax's multimodal model family for generating and editing images, video, and audio through natural-language instructions. This page provides MiniMax H3 Text-to-Video: turn one written prompt into a 4–15-second video at 24 FPS, at either 768p or 2K, with native stereo audio generated with the picture. It accepts text only; use the linked H3 workflows when you need image, video, or audio references.

MiniMax H3 advantage	What it means for you
Selectable 768p or 2K with native stereo audio	Draft faster at 768p, or generate detailed visuals, dialogue, effects, ambience, and music in one pass at 2K—reducing separate upscaling, sound-design, and synchronization stages.
Language-directed Omni-Reference	Across MiniMax H3 workflows, assign images, video, and audio different roles—such as identity, product appearance, motion, camera style, voice, or music—so several sources can guide one coherent result.
V2V transfer and targeted editing	MiniMax H3 can carry over motion or camera language and revise selected visual or audio elements while preserving the rest, giving production teams a path beyond regenerating a shot from scratch.
Delivery-ready timing and framing	Generate 4–15 seconds at 24 FPS in six aspect ratios, so hooks, product reveals, and multi-beat scenes fit cinematic, web, feed, portrait, or vertical placements with less recutting and recropping.

Best Use Cases#

• Advertising and product launches: Use MiniMax H3 for hero shots, product reveals, campaign concepts, and short commercials where camera direction and sound should be designed together.
• Social media content: Use MiniMax H3 to create 9:16 Shorts, Reels, and Stories or 1:1 feed assets with a clear opening hook and platform-ready framing.
• Film and commercial previsualization: Use MiniMax H3 to test a camera move, lighting plan, action beat, scene transition, or temporary soundtrack before committing to production.
• Brand and stylized storytelling: Use MiniMax H3 to explore title sequences, animated posters, character moments, and graphic looks when a written idea needs a polished audiovisual treatment.

How It Works#

1. Write the shot: Tell MiniMax H3 what appears, what changes over time, how the camera moves, how the scene should look, and what should be heard.
2. Choose the delivery format: Select an aspect ratio, a 4–15-second duration, and either 768p or 2k resolution.
3. Generate and refine: The model creates video and stereo audio together. Review motion, text, faces, lip sync, and sound timing, then change one instruction at a time.

Parameters#

The first table lists the controls exposed by the MiniMax H3 Text-to-Video tool on this page.

Parameter	Required	Type	Default	Range / Options	How to choose
prompt*	Yes (*)	String	Example prompt	1–7,000 characters	Give MiniMax H3 a structured brief covering the subject, one main action, camera, setting and lighting, visual style, audio, and intended ending. Clear structure matters more than using the full limit.
aspect_ratio	No	String	16:9	21:9, 16:9, 4:3, 1:1, 3:4, 9:16	Match MiniMax H3 output to the destination: 21:9 for ultra-wide cinematic shots, 16:9 for general video and ads, 4:3 for editorial or retro framing, 1:1 for feeds, 3:4 for portrait products, and 9:16 for vertical social video.
resolution	No	String	768p	768p, 2k	Use 768p for cheaper drafts and faster iteration; use 2k when the shot needs higher detail for large placements or crops.
duration	No	Integer	5	4–15 seconds, in 1-second steps	Use 4–7-second MiniMax H3 clips for one action or fast iteration, 8–10 seconds for a simple change, and 11–15 seconds only when the prompt defines a clear beginning, development, and ending.

• Required field.

The following table summarizes the wider MiniMax H3 model family. Inputs described for first/last-frame and Omni-Reference modes are available through separate workflows, not as controls on this Text-to-Video page.

Core dimension	MiniMax H3
Model	MiniMax-H3
Output duration	MiniMax H3 outputs 4–15 seconds.
Output aspect ratio	First/last-frame mode: follows the original aspect ratio of the input image.<br>Text-to-Video mode: follows the user-selected 21:9, 16:9, 4:3, 1:1, 3:4, or 9:16 ratio.<br>Omni-Reference mode: uses one of those six ratios or Auto, which lets MiniMax H3 choose the output ratio.
Resolution	MiniMax H3 supports two tiers. 768p: for aspect ratios from 16:9 through 9:16, the short side is 768 pixels; for wider output, total resolution is about 1 MP—for example, 21:9 is 1536×672.<br>1440p / 2K: for aspect ratios from 16:9 through 9:16, the short side is 1440 pixels; for wider output, total resolution is about 3.7 MP—for example, 21:9 is 2976×1248.
Output frame rate	MiniMax H3 outputs at 24 FPS.
Output audio	Every MiniMax H3 result includes native stereo audio.
First/last-frame input	Images: 0, 1, or 2; width and height each 256–5760 pixels; aspect ratio from 5:2 to 2:5 (0.4–2.5). With no image input, MiniMax H3 runs in Text-to-Video mode—the tool provided on this page.
Omni-Reference input	Images: up to 9; width and height each 256–5760 pixels.<br>Videos: up to 3; each 2–15 seconds; combined video duration up to 15 seconds; width and height each 256–5760 pixels; aspect ratio from 5:2 to 2:5 (0.4–2.5).<br>Audio: up to 3 clips; each 2–15 seconds; combined audio duration up to 15 seconds. Audio must accompany an image or video and cannot be the only reference.<br>Mixed input: up to 12 files total. With no image, video, or audio input, the workflow becomes Text-to-Video.
Supported input formats	MiniMax H3 accepts video: H.264/AVC or H.265/HEVC; embedded audio: AAC or MP3.<br>Image: JPG, JPEG, PNG, WEBP, HEIC, or HEIF.<br>Audio: WAV or MP3.
Input size limits	Each video: 50 MB; each image: 30 MB; each audio file: 15 MB. There is no separate combined-media limit beyond the per-file limits, but the API request body is limited to 64 MB; URL-based media input is recommended.
Prompt limit	The MiniMax H3 product guide allows up to 7,000 characters. This Text-to-Video deployment currently accepts 1–7,000 characters, as shown in the page-control table above.

Pricing#

MiniMax H3 Text-to-Video pricing depends on resolution and duration:

Resolution	Price per second	5s	10s	15s
768p	$0.10	$0.50	$1.00	$1.50
2K	$0.14	$0.70	$1.40	$2.10

For batches of 1–4 outputs, calculate the total as duration × per-second rate × output count.

Prompting Tips & Examples#

Use this reusable MiniMax H3 prompt structure:

[subject + defining details] + [one action over time] + [camera framing and movement] + [setting + lighting] + [visual treatment] + [dialogue / effects / ambience / music] + [ending frame or constraint]

• Separate motion: Describe subject movement and camera movement independently.
• Show sequence: Use “begins with,” “then,” and “ends on” when the clip has more than one beat.
• Name sound sources: Tell MiniMax H3 who speaks, what makes each sound, the ambience, and whether music should lead or stay subtle.
• Limit competing ideas: One main subject, one primary action, and one coherent style are easier for MiniMax H3 to follow.

Weak prompt

> A luxury watch ad, cinematic and dynamic, with music.

Improved MiniMax H3 prompt

> A brushed-steel automatic watch rests on black volcanic stone. A narrow studio light sweeps across the sapphire crystal as the second hand moves and condensation beads on the case. Begin with an extreme macro, then make a slow 30-degree orbit, ending on the watch face. High-contrast luxury commercial, deep black background. Audio: soft mechanical ticking and one low cinematic pulse; no dialogue.

The improved version gives MiniMax H3 an identifiable subject, timed action, separate camera direction, lighting, finish, sound sources, and a final frame that can be reviewed.

How MiniMax H3 Compares#

Use this MiniMax H3 comparison as a model-family guide; maximum resolution and duration may not be available together, and RunComfy workflows can expose different inputs, resolution tiers, and audio controls.

Model	Resolution	Max duration	Audio	Standout
MiniMax H3	768p or 2K	15s	Native stereo (voice, SFX, music)	Relates text, image, video, and audio references through language for V2V motion transfer and production editing; public weights are planned but have not yet been released.
Hailuo 02	1080p	~10s	None	Strong prompt adherence and physics-focused motion suit gymnastics, dance, product movement, and other silent action shots that will receive audio later in production.
Kling 3.0	Up to 4K	15s	Native audio with lip-sync	Coordinates multi-shot camera changes with multilingual, speaker-assigned dialogue and lip-sync—useful for scripted ads, storyboards, and character-led scenes.
Seedance 2.0	1080p	15s	Joint audio and video	Combines dense image, video, and audio references with joint audiovisual generation and precise lip-sync for identity-sensitive ads, branded stories, and reference-heavy edits.
Veo 3.1	4K	8s	Native dialogue and effects	Pairs prompted dialogue and effects with first/last-frame, reference-image, and scene-extension controls—suited to cinematic transitions and assembled sequences.

For Hailuo 02, the listed maxima are mode-specific: 1080p output is limited to 6 seconds, while 10-second output is available at 512p or 768p.

For MiniMax H3, Omni-Reference and V2V capabilities belong to sibling workflows; the Text-to-Video tool on this page remains prompt-only.

What sets MiniMax H3 apart is the combination: one general-purpose model family that relates text, image, video, and audio, generates native stereo sound, and offers both 768p ($0.10/s) and 2K ($0.14/s). Choose MiniMax H3 when you want to start from a text brief and keep a path to reference-guided creation or editing in sibling workflows. Based on publicly available information, run the same brief through each model before committing a pipeline.

More Models to Try#

If MiniMax H3 is not the right starting point for a project, compare these focused workflows on RunComfy:

• MiniMax H3 Image-to-Video: Start from an image and optionally guide the final frame when composition, identity, or product appearance must stay anchored.
• MiniMax H3 Reference-to-Video: Combine images with optional video and audio when motion, camera style, voice, music, or other source details must carry over.
• Hailuo 02 Image-to-Video: Try an earlier MiniMax generation when you want a focused image-animation workflow.
• Hailuo 2.3 Pro: Animate a still image in 1080p with stable, physics-aware motion, nuanced facial expression, and natural lighting—well suited to polished product, portrait, and character shots.
• Kling 3.0: Animate a start image with an optional end frame, element references, timed shot prompts, and synchronized audio for controlled brand and character sequences.
• Seedance 2.0 Pro: Mix up to nine images, three videos, and three audio clips to keep identity, camera language, and synchronized speech, effects, and music aligned across 4–15-second clips.

Official Resources#

• MiniMax H3 launch post
• MiniMax official website