MiniMax H3 on Segmind: Max, Highspeed and Lite tiers with native audio
MiniMax H3 is MiniMax's omni-modal video model, now on Segmind in three tiers: Max for the hardest prompts, Highspeed for fast everyday turnaround and Lite for low-cost batches. Each turns a prompt, a still or reference images into a short clip at up to 768P with native stereo audio.
MiniMax H3 (Hailuo 3.0) is MiniMax's omni-modal video model: it reads text, image, video and audio as one context and generates video with native stereo audio in the same pass. It reaches Segmind in three tiers — Max for the hardest prompts, Highspeed for fast everyday turnaround and Lite for low-cost batches. Max covers text and image to video; Highspeed and Lite add reference to video. Clips run up to 15 seconds and up to 768P.
Max text to video
Max is the top-quality H3 tier, built for the hardest prompts — fluids, smoke, reflective surfaces and fast action that lighter tiers tend to smear. Write a prompt of up to 7,000 characters that leads with one clear action and a motivated camera move, then pick 480P for a draft or 768P for the final. Clips run 5 to 15 seconds, and each one is generated with native stereo audio in the same pass.
ModelMiniMax Hailuo H3 Max Text to VideoGenerate lifelike 768P text-to-video with native audio.Inside a cryogenics lab, a gloved researcher tips a steel flask of liquid nitrogen across a stainless bench; dense white fog cascades over the edge and the camera drops low to follow it rolling across the floor, swallowing a row of glass beakers as frost blooms up the flask under cold blue light.
Dense volumetric fog billows over the bench and rolls across the floor while the glass beakers stay sharp and reflective, with native audio in sync.
A knob of butter drops into a screaming-hot cast-iron skillet and bursts into a sizzle; a hand tosses in sliced shiitake mushrooms that crackle and brown as steam curls up over a blue gas flame. Sound: butter sizzling and mushrooms crackling in the pan.
Butter hits a screaming-hot skillet and the shiitake slices crackle and brown in rising steam, played back with a synced stereo sizzle.
Use Max text to video from the API
Send a prompt and settings, and get back an MP4.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"minimax-h3-max-text-to-video",
prompt="Inside a cryogenics lab, a gloved researcher tips a steel flask of liquid nitrogen across a stainless bench; dense white fog cascades over the edge and the camera drops low to follow it rolling across the floor, swallowing a row of glass beakers as frost blooms up the flask under cold blue light.",
duration=6,
resolution="768P",
ratio="16:9",
prompt_expansion_mode="balanced",
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/minimax-h3-max-text-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "Inside a cryogenics lab, a gloved researcher tips a steel flask of liquid nitrogen across a stainless bench; dense white fog cascades over the edge and the camera drops low to follow it rolling across the floor, swallowing a row of glass beakers as frost blooms up the flask under cold blue light.",
"duration": 6,
"resolution": "768P",
"ratio": "16:9",
"prompt_expansion_mode": "balanced"
}' | jq -r .request_id)
# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
-H "x-api-key: YOUR_API_KEY"
# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
-H "x-api-key: YOUR_API_KEY"Max image to video
Max image to video animates a still you supply, keeping its composition and subjects while adding the demanding motion the tier is known for. Send a first frame, a last frame, or both — with both frames it interpolates between them — and describe the motion in the prompt. The output takes its aspect ratio from your frame; choose 480P or 768P and 5 to 15 seconds.
ModelMiniMax Hailuo H3 Max Image to VideoAnimate a still image into 768P video with audio.Dense white fog billows from the tilted steel flask and cascades over the edge of the stainless bench, rolling low across the floor and swallowing the row of glass beakers as frost blooms up the flask under cold blue light.
The still cryogenics bench comes alive as dense liquid-nitrogen fog billows from the tilted flask and rolls across the floor while the beakers stay sharp.
The tight crimson rosebud unfurls petal by petal into a full open bloom, the outer petals curling back first as dew beads tremble on them against the dark background.
With both a first frame (the tight bud) and a last frame (the full bloom), the clip interpolates petal by petal between them.
Use Max image to video from the API
Pass your first-frame image URL with the prompt; add a last_frame_image to interpolate to it.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"minimax-h3-max-image-to-video",
prompt="Dense white fog billows from the tilted steel flask and cascades over the edge of the stainless bench, rolling low across the floor and swallowing the row of glass beakers as frost blooms up the flask under cold blue light.",
first_frame_image="https://segmind-resources.s3.amazonaws.com/input/minimax-h3-max-image-to-video-first-frame-44463ae8.jpg",
duration=6,
resolution="768P",
prompt_expansion_mode="balanced",
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/minimax-h3-max-image-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "Dense white fog billows from the tilted steel flask and cascades over the edge of the stainless bench, rolling low across the floor and swallowing the row of glass beakers as frost blooms up the flask under cold blue light.",
"first_frame_image": "https://segmind-resources.s3.amazonaws.com/input/minimax-h3-max-image-to-video-first-frame-44463ae8.jpg",
"duration": 6,
"resolution": "768P",
"prompt_expansion_mode": "balanced"
}' | 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"Highspeed text to video
Highspeed balances quality and speed for everyday content at 768P, with native stereo audio generated alongside the picture. Give it a prompt that names one subject, one clear action and a camera move, and it returns a 4-to-15-second clip in your chosen aspect ratio. MiniMax states a 5-second clip averages about 11 to 13 seconds to generate, which suits high-volume social and explainer work.
ModelMiniMax Hailuo H3 Highspeed Text to VideoFast 768P text-to-video with native stereo audio.Macro shot on a brushed-steel lab bench: gloved hands drop a white effervescent tablet into a tall glass cylinder of clear water. The tablet fizzes hard - a dense column of bubbles and amber foam surges up the glass and spills over the rim, droplets scattering across the steel. The camera eases in on the rising foam.
A 768P macro lab-fizz reaction plays out with native stereo audio and a slow push-in, the overflowing foam reading clearly at thumbnail size.
A long curving line of colored dominoes stands on a pale wood tabletop. The first tile tips and a chain reaction races along the curve, each domino clacking the next flat, until the last one topples a small stack of wooden blocks. The camera tracks low alongside the falling line.
At 10 seconds the model sustains a full domino cascade end to end in one coherent run down the curving line.
Use Highspeed text to video from the API
Send a prompt and settings, and get back an MP4.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"minimax-h3-highspeed-text-to-video",
prompt="Macro shot on a brushed-steel lab bench: gloved hands drop a white effervescent tablet into a tall glass cylinder of clear water. The tablet fizzes hard - a dense column of bubbles and amber foam surges up the glass and spills over the rim, droplets scattering across the steel. The camera eases in on the rising foam.",
ratio="16:9",
duration=6,
resolution="768P",
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/minimax-h3-highspeed-text-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "Macro shot on a brushed-steel lab bench: gloved hands drop a white effervescent tablet into a tall glass cylinder of clear water. The tablet fizzes hard - a dense column of bubbles and amber foam surges up the glass and spills over the rim, droplets scattering across the steel. The camera eases in on the rising foam.",
"ratio": "16:9",
"duration": 6,
"resolution": "768P"
}' | jq -r .request_id)
# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$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"Highspeed image to video
Highspeed image to video grows motion out of a still you upload, at 768P with native audio. Send a first frame, or a first and last frame to interpolate between them, and write the prompt as the motion that should play out. The output follows the frame's aspect ratio and runs 4 to 15 seconds.
ModelMiniMax Hailuo H3 Highspeed Image to VideoAnimate a still image into 768P video with sound.The effervescent tablet fizzes hard: a dense column of bubbles and amber foam surges up through the clear water, boils over the rim of the glass cylinder and spills down onto the brushed-steel bench, droplets scattering across the metal. The gloved hand holds the cylinder steady as the camera eases in on the rising foam.
The calm glass of water erupts into a dense column of fizzing amber foam that boils over the rim, with the hiss in the native audio.
The blacksmith swings the hammer down onto the glowing orange bar on the anvil; sparks burst and scatter across the dark forge as the hot metal rings out. The camera holds on the anvil.
The raised hammer swings down and strikes the glowing bar, throwing a burst of sparks the input still never held, with the metal's ring in the audio.
Use Highspeed image to video from the API
Pass your first-frame image URL with the prompt; add a last_frame_image to interpolate to it.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"minimax-h3-highspeed-image-to-video",
prompt="The effervescent tablet fizzes hard: a dense column of bubbles and amber foam surges up through the clear water, boils over the rim of the glass cylinder and spills down onto the brushed-steel bench, droplets scattering across the metal. The gloved hand holds the cylinder steady as the camera eases in on the rising foam.",
duration=6,
resolution="768P",
first_frame_image="https://segmind-resources.s3.amazonaws.com/input/minimax-h3-highspeed-image-to-video-first-frame-34695de1.jpg",
)
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-highspeed-image-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "The effervescent tablet fizzes hard: a dense column of bubbles and amber foam surges up through the clear water, boils over the rim of the glass cylinder and spills down onto the brushed-steel bench, droplets scattering across the metal. The gloved hand holds the cylinder steady as the camera eases in on the rising foam.",
"duration": 6,
"resolution": "768P",
"first_frame_image": "https://segmind-resources.s3.amazonaws.com/input/minimax-h3-highspeed-image-to-video-first-frame-34695de1.jpg"
}' | 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"Highspeed reference to video
Reference to video builds a fresh 768P shot around media you supply, keeping a character, product or place consistent. Give it up to nine reference images, plus optional reference videos for motion and reference audios for a voice. Name each as 'Image 1' and 'Image 2' in the prompt, describe the action, and clips run 4 to 15 seconds with native audio.
ModelMiniMax Hailuo H3 Highspeed Reference to VideoKeep a character, product or place consistent across AI video.The scientist in Image 1 - magenta safety goggles, teal nitrile gloves - drops a white effervescent tablet into a tall glass cylinder of clear water on a brushed-steel lab bench; it fizzes violently, a dense column of bubbles and amber foam surges up the glass and spills over the rim, droplets scattering across the steel. The camera tightens on the overflowing rim.
The supplied scientist — magenta goggles, teal gloves, silver pixie cut — stays recognizably the same in a brand-new lab scene.
On a concrete rooftop terrace, the woman from Image 1 - short copper-red bob, charcoal wool turtleneck - lifts the oversized round tortoiseshell sunglasses from Image 2 onto her face, settles them, and turns to the camera with a slight smile. The camera arcs slowly around her.
Two separate references fuse into one shot: the same woman now wearing the exact round tortoiseshell sunglasses from the second image.
Use Highspeed reference to video from the API
List your reference image URLs and send the prompt; add reference_videos or reference_audios to steer motion or a voice.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"minimax-h3-highspeed-reference-to-video",
prompt="The scientist in Image 1 - magenta safety goggles, teal nitrile gloves - drops a white effervescent tablet into a tall glass cylinder of clear water on a brushed-steel lab bench; it fizzes violently, a dense column of bubbles and amber foam surges up the glass and spills over the rim, droplets scattering across the steel. The camera tightens on the overflowing rim.",
reference_images=[
"https://segmind-resources.s3.amazonaws.com/input/minimax-h3-highspeed-reference-to-video-ref-scientist-b6f14843.jpg",
],
duration=6,
resolution="768P",
ratio="16:9",
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/minimax-h3-highspeed-reference-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "The scientist in Image 1 - magenta safety goggles, teal nitrile gloves - drops a white effervescent tablet into a tall glass cylinder of clear water on a brushed-steel lab bench; it fizzes violently, a dense column of bubbles and amber foam surges up the glass and spills over the rim, droplets scattering across the steel. The camera tightens on the overflowing rim.",
"reference_images": [
"https://segmind-resources.s3.amazonaws.com/input/minimax-h3-highspeed-reference-to-video-ref-scientist-b6f14843.jpg"
],
"duration": 6,
"resolution": "768P",
"ratio": "16:9"
}' | jq -r .request_id)
# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
-H "x-api-key: YOUR_API_KEY"
# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
-H "x-api-key: YOUR_API_KEY"Lite text to video
Lite is the low-cost tier for exploration and batch work, with strong instruction following at 480P or 768P. Write a detailed prompt — subject, wardrobe, setting, camera move and motion — and it places each named element into one coherent shot of 4 to 15 seconds. MiniMax states a 5-second clip averages about 8 to 9 seconds to generate, so it works as a fast draft-and-iterate engine before a higher-fidelity pass.
ModelMiniMax Hailuo H3 Lite Text to VideoText-to-video drafts and batches at 480P or 768P.A runway model strides down a glossy black catwalk in a floor-length crimson silk gown with a long sweeping train. The fabric billows and snaps with each step, and she pivots hard at the end so the train whips across the polished floor. Overhead spotlights rake the runway.
The runway model strides and pivots while the long crimson train billows and whips across the floor, coherent fabric motion for the full clip.
An empty opera house: a grand marble staircase sweeps past gilded balconies and rows of red velvet seats toward a glowing crystal chandelier. [zoom]
The camera pushes in down the marble staircase toward the chandelier, every named element present and the [zoom] tag obeyed.
Use Lite text to video from the API
Send a prompt and settings, and get back an MP4.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"minimax-h3-lite-text-to-video",
prompt="A runway model strides down a glossy black catwalk in a floor-length crimson silk gown with a long sweeping train. The fabric billows and snaps with each step, and she pivots hard at the end so the train whips across the polished floor. Overhead spotlights rake the runway.",
ratio="16:9",
duration=6,
resolution="768P",
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/minimax-h3-lite-text-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "A runway model strides down a glossy black catwalk in a floor-length crimson silk gown with a long sweeping train. The fabric billows and snaps with each step, and she pivots hard at the end so the train whips across the polished floor. Overhead spotlights rake the runway.",
"ratio": "16:9",
"duration": 6,
"resolution": "768P"
}' | 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"Lite image to video
Lite image to video animates a single still into prompt-driven motion at 480P or 768P. Send a first frame, a last frame, or both to interpolate between them, and write the prompt as the motion, since the subject already exists in the frame. The output follows the frame's aspect ratio and runs 4 to 15 seconds.
ModelMiniMax Hailuo H3 Lite Image to VideoAnimate a still image into 480P or 768P video.The runway model strides forward down the glossy black catwalk; the crimson silk gown billows and snaps with each step and the long train whips across the polished floor as she pivots hard at the end. Overhead spotlights rake the runway.
The posed model begins striding within the first second, the crimson gown and long train sweeping across the glossy floor from one still.
Dark chocolate pours in a glossy ribbon from the tilted copper jug and spreads across the top of the tart, rippling outward and settling as the jug lifts away.
Dark chocolate pours from the tilted copper jug and spreads and ripples across the tart before settling as the jug lifts away.
Use Lite image to video from the API
Pass your first-frame image URL with the prompt; add a last_frame_image to interpolate to it.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"minimax-h3-lite-image-to-video",
prompt="The runway model strides forward down the glossy black catwalk; the crimson silk gown billows and snaps with each step and the long train whips across the polished floor as she pivots hard at the end. Overhead spotlights rake the runway.",
first_frame_image="https://segmind-resources.s3.amazonaws.com/input/minimax-h3-lite-image-to-video-input-frame-08667239.jpg",
duration=6,
resolution="768P",
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/minimax-h3-lite-image-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "The runway model strides forward down the glossy black catwalk; the crimson silk gown billows and snaps with each step and the long train whips across the polished floor as she pivots hard at the end. Overhead spotlights rake the runway.",
"first_frame_image": "https://segmind-resources.s3.amazonaws.com/input/minimax-h3-lite-image-to-video-input-frame-08667239.jpg",
"duration": 6,
"resolution": "768P"
}' | jq -r .request_id)
# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$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"Lite reference to video
Lite reference to video carries a referenced character, product or place into a freshly generated shot at 480P or 768P. Supply up to nine reference images, plus optional reference video for motion and reference audio for a voice, and name them in the prompt. Describe the scene and motion, and clips run 4 to 15 seconds.
ModelMiniMax Hailuo H3 Lite Reference to VideoTurn reference images into consistent character and product videos.The woman from Reference Image 1, in her floor-length crimson silk gown with the long train, strides down a glossy black catwalk. The silk billows and snaps with each step, and she pivots hard at the end so the train whips across the polished floor. Overhead spotlights rake the runway and seated spectators line both sides.
The reference model's exact face and crimson gown carry into a moving 768P catwalk stride and pivot, identity and wardrobe preserved.
The grand opera house interior from Reference Image 1 — its white marble staircase, gilded balconies and rows of red velvet seats — holds steady while the camera glides forward through the empty foyer and tilts up to the glowing crystal chandelier.
The referenced opera-house interior holds its marble staircase, gilded balconies and red velvet seats as the camera glides forward and tilts up.
Use Lite reference to video from the API
List your reference image URLs and send the prompt; add reference_videos or reference_audios to steer motion or a voice.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"minimax-h3-lite-reference-to-video",
prompt="The woman from Reference Image 1, in her floor-length crimson silk gown with the long train, strides down a glossy black catwalk. The silk billows and snaps with each step, and she pivots hard at the end so the train whips across the polished floor. Overhead spotlights rake the runway and seated spectators line both sides.",
reference_images=[
"https://segmind-resources.s3.amazonaws.com/input/minimax-h3-lite-reference-to-video-ref-model-gown-63554dab.jpg",
],
ratio="16:9",
duration=6,
resolution="768P",
)
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-lite-reference-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "The woman from Reference Image 1, in her floor-length crimson silk gown with the long train, strides down a glossy black catwalk. The silk billows and snaps with each step, and she pivots hard at the end so the train whips across the polished floor. Overhead spotlights rake the runway and seated spectators line both sides.",
"reference_images": [
"https://segmind-resources.s3.amazonaws.com/input/minimax-h3-lite-reference-to-video-ref-model-gown-63554dab.jpg"
],
"ratio": "16:9",
"duration": 6,
"resolution": "768P"
}' | 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 | 768P default; 480P also available | Max and Lite (Highspeed is 768P only) | 768P for finals, 480P for faster drafts |
| duration | 4–15 seconds, default 6 (Max starts at 5) | all | a single beat at the low end; raise toward 15 for a longer action |
| ratio | 16:9 default, plus 9:16, 1:1, 4:3, 3:4 and 21:9 (Highspeed reference defaults to adaptive) | text to video and reference to video | image to video has no ratio control — it follows your input frame |
| prompt_expansion_mode | balanced default, quality or disabled | Max text and image to video | quality enriches a sparse prompt; disabled sends your prompt verbatim |
| prompt | up to 7,000 characters | all | lead with one clear action and a motivated camera move |
| reference_images / reference_videos / reference_audios | up to 9 images, 3 videos and 3 audios (2–15s each) | reference to video (Highspeed, Lite) | one clean shot per subject on a plain background, named in the prompt |
Good to know
- These tiers render at 480P and 768P on Segmind and do not output 2K; Highspeed is 768P only.
- Native stereo audio is generated with the picture on the H3 family, and there is no separate audio control on the text and image endpoints. Reference to video can take an audio reference to follow a voice or sound.
- Image to video takes its aspect ratio from the frame you upload, so compose the still with the element that should move already in it.
- Max handles the hardest prompts most cleanly, while Highspeed and Lite trade some fidelity for speed, so very fine texture, dense crowds and tiny legible text are harder on them — keep one clear primary action per clip.
Resources
- Try MiniMax H3 Max Text to Video on Segmind · API reference
- Try MiniMax H3 Max Image to Video on Segmind · API reference
- Try MiniMax H3 Highspeed Text to Video on Segmind · API reference
- Try MiniMax H3 Highspeed Image to Video on Segmind · API reference
- Try MiniMax H3 Highspeed Reference to Video on Segmind · API reference
- Try MiniMax H3 Lite Text to Video on Segmind · API reference
- Try MiniMax H3 Lite Image to Video on Segmind · API reference
- Try MiniMax H3 Lite Reference to Video on Segmind · API reference
- MiniMax Video Generation guide — H3 generation modes, output specs, input requirements, aspect ratios and prompt-length limits
- Create Video Generation Task (V2) API reference — the V2 request shape, the multimodal content array, and the model, resolution, duration and ratio fields
- MiniMax H3: An Open Model Breaking the Boundaries Between Tasks and Modalities — the official H3 launch: unified multimodal context, native stereo audio and target use cases
- MiniMaxAI/MiniMax-H3 model card — the open-weight model card: multimodal understanding and native stereo audio
- #minimax-h3
- #minimax
- #model-launch
- #video-generation
- #native-audio
