Skip to main content
MiniMax H3 · video

MiniMax H3 video from $0.01/s — premium results, built for iteration.

Text to video, image to video and 768P/2K video-to-video — with native audio, reference control and precise motion. From $0.01 per second — from 1 credit per output second. One credit is $0.01; subscription plans can lower your effective rate. On eligible tiers, that is up to 70%+ below MiniMax’s listed per-second price.

  • 1 credit = $0.01
  • 24 fps output
  • 480P → 2K
  • Up to 9 reference images
  • OpenAI-compatible API
MiniMax H3 · 768P · 10sGenerated on Callirra · silent loop

A closer look

The same H3 shot, with and without prompt refinement

These supporting examples use the same resolution and duration. The only difference is whether the original idea was sent directly or first shaped by Context-IR. Compare the motion, sound cues and continuity before choosing the optional refinement step.

Butterfly in morning light

768p · 5s
Original prompt
Prompt
Macro shot of an iridescent butterfly resting on a lavender flower. Its wings open, it lifts off, and the camera gently follows it into warm morning light. Soft wing flutter and meadow ambience.
Context-IR refined
Prompt
integrated_multimodal_description: [Shot 1] Cinematic, extreme close-up macro static shot angled slightly above a vibrant, iridescent blue and purple butterfly resting on a blooming purple lavender spike in the center foreground. The strongly blurred background features a lush meadow scattered with circular golden bokeh. Warm morning sunlight streams from the upper right, highlighting the intricate black veins and powdery texture of the insect's wings. The butterfly slowly opens its wings wide, exposing a brilliant metallic sheen across the scales, before suddenly pushing off the flower. The camera immediately transitions into a smooth tracking shot, gently following the butterfly as it flutters upward and toward the right. The focus stays firmly locked on the insect as it flies into the intense, hazy golden light, while the slender green lavender stem sways back and forth from the launch impact in the lower left corner of the frame.
overall_soundscape: A continuous, soft meadow tone of wind gently brushing through grass and distant, faint bird chirps sits in the background. The moment the butterfly launches, a pronounced, rapid crisp fluttering of delicate wings emerges distinctly in the foreground, continuing to rustle clearly as the insect ascends through the air.
non_diegetic_music: A slow-tempo arrangement featuring a solo acoustic guitar playing sparse, high-register harmonics over a sustained, low-register cello drone, holding a constant, quiet volume with no sudden swells or rhythmic percussion.

The pair is a visual comparison, not a promise that every prompt needs refinement. Your prompt remains unchanged until you choose to apply the suggested version.

Horse on the wet beach

768p · 8s
Original prompt
Prompt
A handheld camera follows a brown horse running along a wet beach. Its hooves splash through shallow water, and it turns toward the lens as sunlight breaks through the clouds. Natural hoofbeats, surf and wind.
Context-IR refined
Prompt
integrated_multimodal_description: [Shot 1] Cinematic, medium wide tracking shot with a slight handheld shake. A muscular, deep brown horse with a dark, wind-blown mane and tail gallops aggressively through the center of the frame. The setting is a sprawling wet beach, the flat, dark grey sand heavily saturated and acting as a mirror to the sky above. Initially bathed in cool, diffuse light from a heavy overcast sky, the scene features rolling white ocean surf continuously crashing in the background. The camera continuously tracks alongside the running animal as its powerful legs propel it forward, large hooves violently striking the shallow seawater and kicking up high arcs of translucent, white spray. As the gallop continues, the thick cloud cover above begins to part, and brilliant, warm golden sunlight breaks through. The bright rays strike the wet sand and the horse's glistening coat, casting sharp, elongated shadows to the left and adding a glowing rim light to its silhouette. The lighting shift dramatically warms the color palette of the beach. At this moment, the horse briefly turns its large, dark head toward the lens, its ears flicking back and its dark eyes catching a golden glint from the newly emerged sun, before it continues charging forward through the glowing, sea-foam-laced water.
overall_soundscape: Loud, rhythmic splashes and heavy, pronounced thuds dominate the foreground as the horse's hooves repeatedly strike the wet sand and shallow water. A continuous, roaring crash of ocean surf rolls in the background, accompanied by the persistent, sweeping hiss and howl of a strong coastal wind.
non_diegetic_music: N/A

The pair is a visual comparison, not a promise that every prompt needs refinement. Your prompt remains unchanged until you choose to apply the suggested version.

Rippled mirror study

768p · 12s
Original prompt
Prompt
A woman examines her reflection in an old mirror. The mirror surface ripples like water when she touches it, then becomes clear again as the camera slowly moves closer. Quiet room tone and a soft glass resonance.
Context-IR refined
Prompt
integrated_multimodal_description: [Shot 1] Cinematic, medium shot, the camera slowly pushes in toward an ornate, freestanding antique vanity mirror with a heavily tarnished silver frame. Standing slightly offset to the left in the foreground is a young woman with pale skin and long, wavy dark brown hair, wearing a loose-fitting ivory silk slip dress. She is deeply examining her own reflection. The room around her is cast in moody shadows, with a single source of cool, silvery moonlight streaming in from an off-screen window to the right, highlighting the sharp contours of her face and catching the dust motes suspended in the quiet air. She leans forward, her shoulders tensing slightly in concentration, and slowly raises her right hand. Her index finger makes contact with the center of the mirror's glass. Instantly, the solid surface yields, rippling outward in concentric circles like liquid water. Her eyes widen and her lips part in a startled gasp as she instinctively pulls her hand back a fraction of an inch. The camera continues its steady push-in as the watery undulations gradually settle, returning the mirror to a perfectly smooth, crystalline clarity that reflects her awe-struck expression.
overall_soundscape: A hushed, steady room tone anchors the background, punctuated by the faint rustle of silk fabric as the woman moves her arm. The moment her finger makes contact with the mirror, a distinct, resonant glass hum clearly reverberates in the foreground, accompanied by a subtle, watery splashing sound. The shimmering resonance slowly fades out into the quiet room tone as she draws a sharp, audible intake of breath.
non_diegetic_music: N/A

The pair is a visual comparison, not a promise that every prompt needs refinement. Your prompt remains unchanged until you choose to apply the suggested version.

Use MiniMax H3 on Callirra

Video Generator

Video Generator

2K resolution supports more sizes.

(prompts written in the format of the gray example generate the best video)
integrated_multimodal_description: [Shot 1] Describe the subject, setting, action and camera movement.

overall_soundscape: Describe ambience and action sounds, or N/A.

non_diegetic_music: Describe background score, or N/A.

H3 Context-IR

If ordinary prompts work poorly for H3, use the dedicated compiler to rewrite it for H3. It uses a small, usage-based credit charge and keeps your original prompt unchanged until you apply the preview.

0 / 7000
Cost 10 creditsRemaining 0 credits
Buy Credits
Latest Generation

Your generated image or video will appear here.

Recent Tasks

Loading...

One credit pool, transparent per-second pricing, and every finished take saved with its full recipe so you can reuse it instantly.

Rate card

Priced per output second

The same clip is charged the same way on every plan, and the price is on screen before you press generate.

Best value
480P$0.01/s

The cheapest video on Callirra: $0.01 per second, 1 credit per second of output.

Landscape output
864 × 480
Credits per second
1
10s take
10 credits · $0.10

Best for:Drafting, animatics, social cutdowns, bulk iteration

Most popular
768P$0.02/s

The sweet spot: vertical and landscape HD at $0.02 per second.

Landscape output
1344 × 768
Credits per second
2
10s take
20 credits · $0.20

Best for:Reels, Shorts, ads and anything you publish

1080P$0.03/s

Full HD for deliverables, still at $0.03 per second.

Landscape output
1920 × 1056
Credits per second
3
10s take
30 credits · $0.30

Best for:Client work, YouTube, paid social

2K$0.10/s

The top tier: 2560 × 1440 output at $0.10 per second, plus video-to-video.

Landscape output
2560 × 1440
Credits per second
10
10s take
100 credits · $1.00

Best for:Hero shots, crops, large-format delivery

Modes

Four ways to use it, one generator

Switch modes in the same panel you already know: the reference uploads, audio track and first/last-frame controls appear when the mode needs them.

Text to video

Describe the shot and get a video with sound. Prompts up to 7,000 characters — camera, lens, blocking, lighting and dialogue all fit.

  • Up to 15s per generation
  • 16:9, 9:16 and 1:1

Image to video

Animate a still: a product shot, a character, a storyboard frame. Keep the subject and change the world around it.

  • Up to 9 reference images; the first 5 are free, and 2K charges 3 credits for each additional image
  • First frame, or first + last frame for exact landings
  • Up to 15s, up to 1080P, 2K on the top tier

Video to video (768P / 2K)

Restyle or extend existing footage on the 768P or 2K path: change the look, the grade or the world while the motion stays.

  • 768P or 2K
  • 4–15s for video-to-video at 768P; 4–15s at 2K
  • Up to 9 reference images alongside the source video

Showcase

What MiniMax H3 looks like on Callirra

Ten takes rendered on this platform, each with the prompt that made it and the settings it was rendered at. Every number under a clip comes from its run log — pixels, frame rate, the real length (H3 quantises to 24 fps frames, so a 5s request lands at 5.17s), the native audio track, Press “Use this setup” to load any of them into the generator above; nothing is submitted until you press generate.

Masters, prompts and the measurement logs live in the MiniMax H3 Video Atlas (CC BY 4.0). The page plays a lighter transcoded copy so a wall of clips does not cost you a download budget.

Framing

Landscape, vertical and square from the same model

One canvas per delivery target — a wide establishing shot, a social-first vertical, and a square for feeds. The aspect ratio is a setting, not a second model, and the recipe on each card is what we rendered at.

Community

Prompts the community published

Fourteen cases from creators who posted both the clip and the prompt, reproduced verbatim and credited to them. Their footage is not re-hosted anywhere on this page — load a prompt into the generator above and make your own version.

Curated in the MiniMax H3 Video Atlas (CC BY 4.0), which only lists prompts their creators actually published: a prompt reconstructed by watching a video describes the output, not the prompt, and is not passed off as anyone’s work here.

Built for

Made for how you actually ship

Indie filmmakers

Block out coverage at 480P, then finish the selects at 1080P without changing tools.

Ads & marketing

Reference the product, keep it consistent across takes, and re-render one shot instead of the whole spot.

Social creators

9:16 vertical with native audio, priced per second so a week of posts stays affordable.

Developers

One OpenAI-compatible endpoint, billed per output second, sharing the same credit pool as the UI.

API

MiniMax H3 through one endpoint

OpenAI-compatible, billed per output second, sharing the same credit pool as the browser generator. Submit a job, poll it, download the file.

Workflow

From a sentence to a finished take

Free plan · no card required

18 credits your first week, 15 every week after

A new account starts with 18 credits — the 15-credit weekly grant plus a 3-credit welcome bonus — and every week after that resets to 15. On MiniMax H3 that is 18 seconds of 480P video in week one, or 9 seconds at 768P.

  • 480P: 18s in week one, 15s every week after
  • 768P: 9s in week one, 7s every week after
  • 1080P: 6s in week one
  • Paid plans download without a watermark; free-tier downloads carry the Callirra mark

Credits reset every Monday (UTC) and never roll over. Failed generations are refunded automatically.

Generate with free credits

Why Callirra

A cheap model is only useful if the whole pipeline is cheap

Video from $0.01 a second

480P costs 1 credit per second of output and 768P 2 credits. The price is per output second, shown before you generate, so a 10-second take is a number you already know.

One credit pool for images and video

No second wallet and no separate MiniMax account: the same credits that render GPT Image 2.5 stills render H3 video, and every plan spends them at the same rate.

Iterate at 480P, deliver at 1080P

480P is 1 credit per output second, so you can try several takes cheaply before finishing the selected shot at 1080P.

Reference control that holds up

Up to 9 reference images and first/last-frame landing in the same generator — the first 5 images cost nothing.

Failed generations cost nothing

A job that fails is refunded automatically to the pool it came from, so you only spend credits on output you can actually download.

Specs

MiniMax H3 specifications

FAQ

Questions before you spend a credit

Per second of output: $0.01 at 480P, $0.02 at 768P, $0.03 at 1080P and $0.10 at 2K — 1/2/3/10 credits per second, and one credit is one cent. A 10-second 768P clip is 20 credits ($0.20).

Start with a video, not a sales call

10s at 768P costs 20 credits. A new account gets 18 to start, so the first take is on us — then you decide whether the rate works for your volume.

Open the MiniMax H3 generator

Callirra is an independent platform. MiniMax H3 is served through Callirra’s own interface and is not affiliated with, endorsed by, or sponsored by MiniMax. Model names and trademarks belong to their respective owners.

Page artwork was produced on Callirra: the studio backdrop and paper texture with GPT Image 2.5, and the hero loop with MiniMax H3 at 768P. The showcase and community sections above are separate from it — those are the takes and prompts described in their own captions.