MiniMax H3 Optimised: text, image and reference to video with native sound
MiniMax H3 Optimised serves MiniMax's H3 video model on a faster inference stack. Make 5-to-15-second clips with native dialogue, sound and ambience in one pass — from a text prompt, a first frame, or reference images and a reference video.
MiniMax H3 Optimised serves MiniMax's H3 (Hailuo 3.0) video model on an optimised inference stack for faster turnaround at matched quality. It returns a clip that already has sound — dialogue, effects and ambience generated with the picture and timed to the action, with lip movement matching speech. Every clip runs any whole length from 5 to 15 seconds at 24 fps, from 480p up to 2k. It ships in three modalities — text to video, image to video and reference to video.
Text to video
Text to video needs only a written prompt, up to 7,000 characters. Write it like a shot brief — camera move, then subject, then action, then scene — and spell out the spoken line and the sounds you want, because the audio is produced with the picture. Its strongest trick is native dialogue and synced sound in a single pass. Pick an aspect ratio for where the clip will run: 16:9, 9:16, 1:1, 4:3, 3:4 or 21:9.
ModelMiniMax H3 Optimised Text to VideoText-to-video with native dialogue and sound, up to 2k.On the footplate of a steam locomotive pounding along an open track, a soot-streaked driver in a striped cap and grey overalls leans out of the cab window, looks back along the train and says clearly, 'Right, she's holding pressure - take us up the bank.' The firebox throws an orange glow across his face, white steam streams past, and a moment later the whistle lets out a long shriek as the wheels hammer the rails. Medium shot from alongside the moving cab.
The steam-locomotive driver speaks his English line with his lips in sync and the whistle shrieks on cue, all in one pass with native stereo audio.
Medium close-up from across the counter, the camera holding steady. A sushi chef in a white jacket and black bandana presses a piece of salmon nigiri between his palms, sets it down on a dark lacquer board, then looks up to the camera and says clearly in Japanese, 'どうぞ、召し上がれ。' Behind him a cedar counter and a row of glass fridges; the knife clinks on the board and low restaurant chatter fills the room.
The chef presses the nigiri, then looks up and delivers a Japanese line with his mouth clearly forming the words over quiet restaurant ambience.
A steady push-in from a wide frame to a close shot of the hands. At a wooden handloom strung with indigo and cream threads, a weaver in a linen apron throws the wooden shuttle through the warp, pulls the reed toward her to beat the weft tight, then throws the shuttle back the other way as a striped band of cloth grows row by row on the loom. Daylight from a side window catches the lint in the air; the loom clacks and the treadles thump in a steady rhythm.
One unbroken 10-second push-in travels from the wide handloom to the weaver's hands as the striped cloth grows on the warp, the loom clacking in the audio.
Use text to video from the API
Send the prompt and settings and get back an MP4 with its audio track.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"minimax-h3-optimised-text-to-video",
prompt="On the footplate of a steam locomotive pounding along an open track, a soot-streaked driver in a striped cap and grey overalls leans out of the cab window, looks back along the train and says clearly, 'Right, she's holding pressure - take us up the bank.' The firebox throws an orange glow across his face, white steam streams past, and a moment later the whistle lets out a long shriek as the wheels hammer the rails. Medium shot from alongside the moving cab.",
resolution="768p",
duration=5,
aspect_ratio="16:9",
quality="high",
enhance_prompt=True,
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/minimax-h3-optimised-text-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "On the footplate of a steam locomotive pounding along an open track, a soot-streaked driver in a striped cap and grey overalls leans out of the cab window, looks back along the train and says clearly, '\''Right, she'\''s holding pressure - take us up the bank.'\'' The firebox throws an orange glow across his face, white steam streams past, and a moment later the whistle lets out a long shriek as the wheels hammer the rails. Medium shot from alongside the moving cab.",
"resolution": "768p",
"duration": 5,
"aspect_ratio": "16:9",
"quality": "high",
"enhance_prompt": true
}' | jq -r .request_id)
# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
-H "x-api-key: YOUR_API_KEY"
# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
-H "x-api-key: YOUR_API_KEY"Image to video
Image to video animates a still you supply. Upload a first frame and describe what should move, the camera, and any dialogue or sound; the clip opens on your image and follows its aspect ratio, so there is no separate aspect-ratio control. Add an optional last frame to drive a start-to-end transition. It carries the same native dialogue and sound as the rest of the family.
ModelMiniMax H3 Optimised Image to VideoAnimate a first frame into video with lip-synced sound.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 uploaded steam-locomotive still is the opening frame, then the driver turns to camera and calls his line in sync while the whistle shrieks and the engine pounds forward.
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 opening frame matches the uploaded still, and he delivers a Japanese line with his mouth forming the words over quiet restaurant ambience.
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 closed bud supplied as the first frame and resolves onto the fully open bloom supplied as the last frame, unfurling in between.
Use image to video from the API
Pass the first-frame image URL with the prompt; add last_frame_image for a transition.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"minimax-h3-optimised-image-to-video",
image="https://segmind-resources.s3.amazonaws.com/input/minimax-h3-optimised-image-to-video-first-frame.jpg",
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.",
resolution="768p",
duration=5,
quality="high",
enhance_prompt=True,
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/minimax-h3-optimised-image-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"image": "https://segmind-resources.s3.amazonaws.com/input/minimax-h3-optimised-image-to-video-first-frame.jpg",
"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.",
"resolution": "768p",
"duration": 5,
"quality": "high",
"enhance_prompt": true
}' | jq -r .request_id)
# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
-H "x-api-key: YOUR_API_KEY"
# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
-H "x-api-key: YOUR_API_KEY"Reference to video
Reference to video builds a clip around media you supply, and needs at least one reference. Give it up to four reference images, named Image 1 to Image 4 in the prompt, to fix a character, product or scene. Add one optional reference video, named Video 1, to borrow a camera move or motion. Name each reference and its job in the prompt, then describe the action and sound as usual.
ModelMiniMax H3 Optimised Reference to VideoKeep characters and motion from reference images and video.Image 1 is the locomotive driver. On the footplate of a steam locomotive pounding along an open track, he leans out of the cab window, looks back along the train and says clearly, 'Right, she's holding pressure - take us up the bank.' Follow the camera movement from Video 1. The firebox throws an orange glow across his face, white steam streams past, and a moment later the whistle lets out a long shriek as the wheels hammer the rails. Medium shot from alongside the moving cab.
The generated driver matches the reference portrait and the camera tracks alongside the engine exactly as the reference clip, the spoken line lip-synced and the whistle on the beat.
A low tracking shot that follows the toy across a sunlit wooden floor. Image 1 is the boy in blue-striped pyjamas with tousled brown hair; Image 2 is the red tin fire truck with a white ladder. He kneels, pushes the fire truck forward with one hand and calls out a bright 'wee-ooo' as it rolls; the brass bell on the truck dings twice. Morning light falls from a window at the side. Sound: the truck's wheels rumbling on the boards, the small bell, the boy's voice. The truck rolls to a stop as he grins after it.
Both the boy (Image 1) and the red fire truck (Image 2) appear together in one shot, each recognisably matching its upload.
A slow quarter-turn arc around a teal high-top sneaker standing on a round acrylic plinth in a bright product studio. Image 1 is the sneaker: teal canvas upper, a single orange side stripe, white laces and an amber gum sole. A hand reaches in and turns it a quarter spin so the orange stripe and gum sole come round to the lens, then lets it settle. Even studio light, pale seamless background. Sound: the faint squeak of rubber on acrylic and a low studio room tone. The sneaker rests three-quarters to camera.
The teal high-top's colourway, orange stripe and gum sole hold through a quarter-turn arc, proving a packshot carries across a moving shot.
Use reference to video from the API
List the reference image URLs, add a reference video URL for a camera move, and send the prompt.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"minimax-h3-optimised-reference-to-video",
prompt="Image 1 is the locomotive driver. On the footplate of a steam locomotive pounding along an open track, he leans out of the cab window, looks back along the train and says clearly, 'Right, she's holding pressure - take us up the bank.' Follow the camera movement from Video 1. The firebox throws an orange glow across his face, white steam streams past, and a moment later the whistle lets out a long shriek as the wheels hammer the rails. Medium shot from alongside the moving cab.",
reference_images=[
"https://segmind-resources.s3.amazonaws.com/input/minimax-h3-optimised-r2v-driver-ref.jpg",
],
reference_video="https://segmind-resources.s3.amazonaws.com/input/minimax-h3-optimised-r2v-camera-ref.mp4",
resolution="768p",
duration=5,
aspect_ratio="16:9",
quality="high",
enhance_prompt=True,
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/minimax-h3-optimised-reference-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "Image 1 is the locomotive driver. On the footplate of a steam locomotive pounding along an open track, he leans out of the cab window, looks back along the train and says clearly, '\''Right, she'\''s holding pressure - take us up the bank.'\'' Follow the camera movement from Video 1. The firebox throws an orange glow across his face, white steam streams past, and a moment later the whistle lets out a long shriek as the wheels hammer the rails. Medium shot from alongside the moving cab.",
"reference_images": [
"https://segmind-resources.s3.amazonaws.com/input/minimax-h3-optimised-r2v-driver-ref.jpg"
],
"reference_video": "https://segmind-resources.s3.amazonaws.com/input/minimax-h3-optimised-r2v-camera-ref.mp4",
"resolution": "768p",
"duration": 5,
"aspect_ratio": "16:9",
"quality": "high",
"enhance_prompt": true
}' | jq -r .request_id)
# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
-H "x-api-key: YOUR_API_KEY"
# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
-H "x-api-key: YOUR_API_KEY"Shared settings
| Setting | Value | Members | When to use it |
|---|---|---|---|
| resolution | 480p, 768p, 1080p, 2k (default 768p) | all | 768p is the only natively generated tier; 1080p and 2k are upscaled, so reach for them only when you need more pixels |
| duration | 5–15 whole seconds (default 5) | all | 5 for a single beat; raise toward 15 for a short sequence, adding continuity cues |
| quality | high or low (default high) | all | high for finals; low renders faster and looks nearly identical for drafts |
| enhance_prompt | on by default | all | leave on to expand a terse brief; turn off to send a finished prompt verbatim |
| aspect_ratio | 16:9, 9:16, 1:1, 4:3, 3:4, 21:9 (default 16:9) | text to video, reference to video | image to video has no control here — the clip follows the first frame's shape |
| image, last_frame_image | first frame required; last frame optional | image to video | the clip opens on the first frame; add a last frame for a start-to-end transition |
| reference_images, reference_video | up to 4 images; one optional video | reference to video | images fix a character, product or scene (Image 1–4); the video supplies a camera move (Video 1) |
| seed | any integer | all | set a value to reproduce a clip |
Good to know
- Audio is generated with the picture, but spoken dialogue is harder than ambient sound. Keep a line to one short, clear sentence at 5 seconds so the lip-sync holds, and add continuity cues on 12–15 second takes.
- Only 768p is generated natively. 1080p is upscaled from 540p and 2k is upscaled from 768p, so reach for them only when you need more pixels.
- Rendered on-screen text is a weak area. Avoid prompts that rely on legible text inside the frame.
- 4k is available on the API only and returns HEVC, which most browsers play as audio only, so it is left out of the playground resolution list.
- Image to video takes its aspect ratio from the first frame, and reference to video needs at least one reference image or video.
Resources
- Try MiniMax H3 Optimised Text to Video on Segmind · API reference
- Try MiniMax H3 Optimised Image to Video on Segmind · API reference
- Try MiniMax H3 Optimised Reference to Video on Segmind · API reference
- MiniMax H3: An Open Model Breaking the Boundaries Between Tasks and Modalities — MiniMax's own H3 announcement: the omni-modal design that reads text, image, video and audio as one context, native stereo audio generated with the picture, and the task-generalization philosophy behind the model
- MiniMax H3 Model - Native Multimodal Video Generation (Hailuo) — MiniMax's Hailuo product page for H3: creation modes (text-to-video, first/last-frame, omni-reference), supported aspect ratios, 5-15s at 24fps, native stereo sound, a step-by-step how-to-prompt workflow and an FAQ
- #minimax-h3-optimised
- #minimax
- #model-launch
- #video-generation
- #native-audio
