MiniMax H3 Optimised Image to Video

MiniMax H3 Optimised Image to Video

MiniMax H3 Optimised Image to Video turns a single still into a short, sound-complete clip. Upload a first frame, describe what should move, and the model animates it with dialogue, sound effects and ambience generated in the same pass.

What is MiniMax H3 Optimised Image to Video?

MiniMax H3 is MiniMax's third-generation Hailuo video model: an omni-modal system that reads text, images, video and audio as one context and returns video with native stereo sound. This image-to-video endpoint takes your first frame and a prompt and produces a 5 to 15 second clip at 24 fps whose opening frame matches your still. Served on an optimised inference stack, it keeps that quality with faster turnaround. The output follows the first frame's aspect ratio, so framing is set by the image you upload.

Key Features

• •Animate a first frame, with an optional last frame for a controlled start-to-end transition.
• •Native synchronized stereo audio: spoken lines land on mouth movements and effects land on the action.
• •Literal, director-style prompt adherence for camera, subject and action.
• •5 to 15 second clips in whole-second steps at a film-standard 24 fps.
• •Resolutions from 480p to 2k, where 768p is native and 1080p and 2k are upscaled.
• •A quality switch, an optional prompt-enhance pass, and multilingual dialogue with lip-sync.

Best Use Cases

Bring a product shot, character render, poster or concept frame to life; animate a fixed opening frame into a talking-character scene; build social and short-form spots; and previsualize from art you already have. The optional last frame gives a clean A-to-B transition, and 24 fps output cuts straight into an editing timeline.

Prompt Tips and Output Quality

Upload a clean, uncluttered first frame and name exactly what should move, the camera move, and any spoken line or sound. Keep dialogue to one short, clear line for the tightest lip-sync. In testing, a 768p five-second clip returned 1344x768 with stereo audio, the opening frame matching the uploaded still and the dialogue and scene sounds landing on the action. Avoid asking for legible on-screen text, which is weaker here.

How to Use MiniMax H3 Optimised Image to Video

MiniMax H3 Optimised Image to Video animates a still you provide. You supply two things: a first frame and a prompt describing the motion, dialogue and sound you want, and the model returns a clip whose opening frame matches your image.

Preparing your first frame

Upload a clean, well-composed still. The output follows this frame's aspect ratio, so there is no separate aspect-ratio control — crop the image to the shape you want, whether widescreen, vertical or square. Leave room for the motion you plan to describe, and keep the frame uncluttered when you need tight continuity.

Writing the prompt

Describe what in the frame should move, then the camera move, then any spoken line and the sounds of the scene. Because audio is generated natively, spell out the dialogue and the effects you expect to hear. Keep spoken lines short and clear — one line gives the most accurate lip-sync. Avoid asking for legible on-screen text.

Optional last frame

Add a last frame to drive a start-to-end transition: the clip begins on your first frame and resolves onto the last one. Use matching framing for a smooth result, or leave it empty for first-frame-only motion.

Recommended Settings

• •resolution: 768p — the natively generated tier; choose 1080p or 2k only when you need more pixels, since both are upscaled.
• •duration: 5 seconds for a single beat or tight dialogue; up to 15 for longer sequences, where continuity cues help.
• •quality: high for finals; low renders faster and looks nearly identical for drafts.
• •enhance_prompt: on by default; turn it off to send a finished prompt verbatim.
• •seed: set a value to reproduce a result.

In testing, a 768p five-second render returned a clean 1344x768 clip at 24 fps with stereo audio: the opening frame matched the uploaded still, the subject spoke a lip-synced line, and the scene sounds landed on the action.

Examples to try

Each example ran end to end at the settings on its card — 768p, five seconds, high quality — so every clip is 1344x768 at 24 fps with native stereo sound, and the opening frame is the still that was uploaded.

• •Native dialogue and sound in one pass (default). Prompt: "The soot-streaked driver leaning from the cab window turns his head back toward the camera and calls out clearly, "Right, she's holding pressure - take her up the bank!" The firebox flames flicker orange across his face, white steam billows thicker past the boiler, and the locomotive pounds forward along the rails. A long steam-whistle shriek rings out and the wheels hammer the track. Camera holds alongside the moving cab." The spoken line is lip-synced and the whistle lands on cue; keep dialogue to one short line at 5s so the sync holds.
• •Dialogue in a second language with lip-sync. Prompt: "Behind the cypress counter, the sushi chef presses the salmon nigiri once with two fingers, then raises his eyes to the camera and says clearly in Japanese, "どうぞ、召し上がれ。" The camera holds steady at the counter. Quiet restaurant ambience and a faint knife tap on the board." The chef's mouth forms the Japanese words over room ambience; put the exact foreign line in quotes and keep it to one short sentence.

First-and-last-frame examples

Supply a last frame and the model works from two stills — the first frame it opens on and the last frame it resolves to; upload each to its matching field.

• •Optional last frame for a start-to-end transition. Prompt: "The closed burgundy tulip bud swells and its ruffled petals peel open one by one until the flower stands fully bloomed, facing the camera. The camera holds in a tight macro. A faint room tone and the dry rustle of petals." The clip opens on the bud and resolves onto the open bloom; keep the two frames in matching framing for the smoothest result.

Built for more: try "upload a portrait and have them deliver one line to camera", or "give a first and last frame — a shutter closed, then open — for a clean reveal."

Does it generate sound?

Yes — dialogue, effects and ambience are produced with the picture and timed to it.

Does it keep my input image?

The output's first frame matches the still you upload, as closely as native H3.

Can I control the ending?

Add an optional last frame for a start-to-end transition.

How long can clips be?

Any whole length from 5 to 15 seconds at 24 fps.

What resolution should I use?

768p is native; 1080p and 2k are upscaled, so choose them only when you need more pixels.