Skip to content
Creative Studio, runs on our GPU

Cinematic video with sound Runs on our GPU

AI video generator with sound

Write the scene, get the shot, sound included. Film-style clips up to 20 s with sound effects, voices and music, from one prompt.

Create

22 credits · about 22 min on our GPU with MiniMax H3 · Final · up to 12 sQuickest: LTX-2.5 · Preview · up to 20 s, 1 credit

What it does

You describe a scene, its camera moves and its soundscape; the model renders a short film-style clip with matching audio: effects, dialogue lines and music. MiniMax H3 gives the most cinematic picture; LTX-2.5 is several times faster and cheaper for the same length.

What you can do

9 real jobs, run on our GPU

Each example is a real run on our GPU: what went in, the exact instruction, and the result exactly as the model made it. Pick one to see it, then start from the same instruction.

Battlefield: soldiers, jets and radio comms

What went in

Only the instruction: no file.

What we asked

integrated_multimodal_description: [Shot 1] Live-action, cinematic 2.39:1 anamorphic, gritty war-film look, dusty golden-hour light cutting through smoke. Medium-wide shot of elite special-forces soldiers in modern multicam gear advancing through a ruined urban street under heavy fire. Tracer rounds streak past; one soldier drops behind a crumbling wall and returns fire with controlled bursts. [Shot 2] At 00:04.000, hard cut to low tracking shot following a second soldier as he sprints, radio handset pressed to his ear. [Shot 3] At 00:07.500, cut to dynamic low-angle as a pair of modern stealth fighter jets scream overhead, afterburners glowing, dropping flares. Soldiers look up, then continue advancing. Camera pushes in slowly on the lead soldier's determined face. overall_soundscape: Intense battlefield ambience — distant explosions, close gunfire cracks, ricochets, jet roar fly-by, boots on rubble, heavy breathing. Radio static and clear military comms (S1, urgent male voice): "Eagle One, this is Ground Lead — request immediate close air support, grid Alpha-7, danger close!" non_diegetic_music: Low, pulsing hybrid score — deep brass hits, tense electronic pulses, rising strings that swell under the jet pass.

What came out, untouched

First frame of the clip: Battlefield: soldiers, jets and radio comms
Play the full clip in the gallery

32 credits ($3.20) · MiniMax H3 · Final · 13-15 s

Jet dogfight and missile

What went in

Only the instruction: no file.

What we asked

integrated_multimodal_description: [Shot 1] Live-action, ultra-cinematic aerial photography, high-contrast blue sky and white contrails, anamorphic lens flares. Extreme wide aerial shot of two modern 5th-generation fighter jets locked in a high-G dogfight — one banking hard left, the other rolling inverted to gain position. [Shot 2] At 00:03.500, cut to cockpit POV with HUD elements and rapid head movement as the pilot looks over his shoulder. [Shot 3] At 00:06.000, wide side tracking shot as a ground-based S-300 missile launches from a distant mountain ridge, white smoke trail racing upward toward the jets. One jet breaks hard and deploys a cloud of flares. Camera arcs around the lead jet as it pulls into a vertical climb. overall_soundscape: Roaring afterburners, wind rush at high altitude, cockpit radio chatter with clipped military tone (S1): "Missile launch! Break right, flares now!" Missile motor ignition whoosh, flare pop and burn. non_diegetic_music: Aggressive modern action score — driving percussion, distorted synths, rising brass stabs that peak on the missile launch.

What came out, untouched

First frame of the clip: Jet dogfight and missile
Play the full clip in the gallery

32 credits ($3.20) · MiniMax H3 · Final · 13-15 s

Sports car and police chase

What went in

Only the instruction: no file.

What we asked

integrated_multimodal_description: [Shot 1] Live-action, high-octane cinematic car-chase style, wet night city streets reflecting neon, anamorphic bokeh. Low tracking shot racing beside a matte-black high-performance sports car as it blasts through an intersection, tires smoking. [Shot 2] At 00:04.000, cut to rear three-quarter pursuit angle as multiple police interceptors with flashing blue-red lights close in, sirens wailing. The sports car drifts hard around a tight corner, sparks flying from the undercarriage. [Shot 3] At 00:08.000, dynamic low front tracking shot as the car accelerates away under a bridge, police lights painting the wet asphalt. Camera shakes slightly with speed. overall_soundscape: Roaring V8 / high-revving engine, tire screech, police sirens Doppler effect, radio chatter "Suspect vehicle is a black sports car, do not lose him!", wet road spray, distant city traffic. non_diegetic_music: Fast electronic rock / hybrid chase score — aggressive synth bass, driving drums, electric guitar stabs that intensify with every near-miss.

What came out, untouched

First frame of the clip: Sports car and police chase
Play the full clip in the gallery

32 credits ($3.20) · MiniMax H3 · Final · 13-15 s

Snowy mountains under the galaxy

What went in

Only the instruction: no file.

What we asked

integrated_multimodal_description: [Shot 1] Live-action, epic nature cinematography, crystal-clear night, deep blue and silver tones. Ultra-wide establishing shot of jagged snow-covered mountain peaks under a vast, dense Milky Way galaxy stretching from horizon to horizon. Soft aurora green briefly dances on the far ridge. [Shot 2] At 00:05.000, slow majestic crane up and push forward through light falling snow, revealing a lone figure standing on a ridge looking upward. Stars slowly rotate as the camera continues its gentle ascent. overall_soundscape: Profound silence broken only by soft wind across snow, distant ice cracks, faint high-altitude hush. Occasional quiet footstep crunch if the figure moves. non_diegetic_music: Sparse, emotional ambient score — deep sustained strings, subtle choir pads, single piano notes that bloom under the galaxy reveal.

What came out, untouched

First frame of the clip: Snowy mountains under the galaxy
Play the full clip in the gallery

32 credits ($3.20) · MiniMax H3 · Final · 13-15 s

Rainforest clearing

What went in

Only the instruction: no file.

What we asked

integrated_multimodal_description: [Shot 1] Live-action, poetic nature cinema, soft golden late-afternoon light filtering through canopy. Wide shot of a dense, vibrant rainforest clearing with moss-covered rocks, cascading small waterfall, and rich green ferns. A solitary person in simple outdoor clothing sits quietly on a large flat stone, back partially to camera, watching the water. [Shot 2] At 00:05.000, slow gentle push-in and slight arc around the person as a single leaf drifts down and lands on the water surface. The person remains still, breathing calmly. Light rays shift subtly through the trees. overall_soundscape: Rich natural soundscape — flowing water, distant birdsong, soft insect hum, gentle leaf rustle, occasional distant animal call. Quiet breathing of the person. non_diegetic_music: Minimalist ambient score — soft acoustic guitar arpeggios, warm pad tones, subtle nature-inspired textures that remain understated.

What came out, untouched

First frame of the clip: Rainforest clearing
Play the full clip in the gallery

30 min 42 s on our RTX 4060 32 credits ($3.20) · MiniMax H3 · Final · 13-15 s

Street football at golden hour

What went in

Only the instruction: no file.

What we asked

integrated_multimodal_description: [Shot 1] Live-action, cinematic sports film, warm golden-hour sunlight cutting between colorful buildings in Rio de Janeiro. Wide low-angle shot of a narrow Brazilian street, graffiti-covered walls, hanging electrical wires, and distant favelas. A young Brazilian footballer in a worn jersey dribbles a football skillfully across cracked pavement, approaching a makeshift goal between two buildings. [Shot 2] At 00:05.000, dramatic slow motion as the player plants his foot and prepares to strike. Camera rapidly arcs 180 degrees around him, tilting upward from ground level as his leg swings forward. Dust rises, sunlight flares across the lens, and the ball launches toward the goal with explosive power. overall_soundscape: Authentic Rio street atmosphere — distant Portuguese chatter, children shouting, rhythmic footsteps, football bouncing against concrete, birds, and a sharp powerful kick echoing between buildings. non_diegetic_music: Cinematic Brazilian percussion blended with rising orchestral tension — deep rhythmic drums, subtle samba-inspired beats, swelling bass, and a dramatic impact accent as the ball leaves his foot.

What came out, untouched

First frame of the clip: Street football at golden hour
Play the full clip in the gallery

20 min 35 s on our RTX 4060 22 credits ($2.20) · MiniMax H3 · Final · up to 12 s

Secret tropical paradise travel reel

What went in

Only the instruction: no file.

What we asked

Generate a breathtaking 12-second cinematic travel Reel revealing a secret tropical paradise. Open with a close-up of a hand pulling back a curtain in a dark room. As the curtain opens, transition seamlessly into an aerial drone shot of impossibly turquoise water, a tiny white-sand island, dramatic green cliffs, and a luxury villa perched above the ocean. Use a fast camera push-in, realistic waves, moving palm trees, golden-hour sunlight, natural colors, and premium travel-commercial cinematography. End with a spectacular overhead reveal. First-frame text: “Would you disappear here for 7 days?” Vertical 9:16, photorealistic, immersive natural audio, no logos, 12 seconds, visually compelling in the first second.

What came out, untouched

First frame of the clip: Secret tropical paradise travel reel
Play the full clip in the gallery

22 min 26 s on our RTX 4060 22 credits ($2.20) · MiniMax H3 · Final · up to 12 s

Jet dogfight, LTX-2.5 (compare with H3)

What went in

Only the instruction: no file.

What we asked

Ultra-cinematic aerial dogfight, bright high-altitude light and long contrails. Extreme wide shot of two modern stealth fighters locked in a high-G turning fight, one banking hard while the other rolls inverted. Quick cut to cockpit POV with HUD glow as the pilot snaps his head. A ground-based S-300 missile launches from a distant ridge, white smoke trail streaking upward; the lead jet breaks hard and dumps a brilliant cloud of flares. Camera arcs tightly around the climbing jet. Audio: thunderous afterburners, high-altitude wind, urgent radio "Missile launch! Break right, flares now!", missile ignition roar and flare pops under a driving action score of heavy percussion and rising brass.

What came out, untouched

First frame of the clip: Jet dogfight, LTX-2.5 (compare with H3)
Play the full clip in the gallery

7 min 24 s on our RTX 4060 8 credits ($0.80) · LTX-2.5 · Balanced · 13-20 s

Cat CEO comedy reel

What went in

Only the instruction: no file.

What we asked

Create a 12-second ultra-realistic, funny Instagram Reel featuring a fluffy orange cat acting like a billionaire CEO in a luxury skyscraper office. First second: the cat slams a tiny paw onto a desk as the caption appears: “POV: Your boss is a cat.” The cat confidently reviews charts, fires a mouse-shaped employee, and jumps onto a giant leather chair. Final second: the cat accidentally knocks every document off the desk and freezes with an innocent expression. Cinematic lighting, realistic fur, expressive eyes, rapid comedic cuts, perfect sound effects, vertical 9:16, no logos, seamless ending that loops back to the opening. Make it instantly understandable without dialogue.

What came out, untouched

First frame of the clip: Cat CEO comedy reel
Play the full clip in the gallery

4 min 17 s on our RTX 4060 5 credits ($0.50) · LTX-2.5 · Balanced · up to 12 s

Works well for

  • Instagram Reels and YouTube Shorts openers
  • Faceless channel b-roll with sound
  • Travel and nature mood clips
  • Action and sci-fi concept shots for pitches
  • Storyboard tests before a real shoot
  • Music-video and trailer ideas

Check before you rely on it

  • Slow: MiniMax H3 Final takes 20-32 minutes per clip on this GPU; you get an email when it is ready.
  • No readable text in the picture: captions asked for in the prompt do not come out legible.
  • LTX-2.5 draws fast jets and small moving objects soft and smeared (see the dogfight comparison).
  • Real places, signs and number plates can appear with garbled lettering.
  • Sexual content, hate and holy or religious places and symbols are refused by the filter; refused jobs are refunded.

Get the best result

The prompt and the model decide the output; bigger models and settings cost more credits. These notes come from what worked, and what did not, in our runs.

Do

  • Write one dominant camera move and one emotional arc per clip; every good trial clip had exactly that.
  • Describe the sound: name the effects, the music and any spoken line in quotes. Both models generate audio from those words.
  • Keep MiniMax H3 at 8-12 s: its shot timestamps assume about 10 s, and the 12 s option costs 22 credits against 32 for 15 s.
  • Try a cheap LTX-2.5 Preview first when you are unsure of a prompt, then render the keeper on H3 Final.

Avoid

  • Ask for on-screen captions or titles: the trial prompts that asked for caption text did not get readable text. Add titles in your editor.
  • Stack many scene changes: every good trial clip used two or three shots with timestamps.
  • Write the frame shape only in the prompt: choose the 9:16 setting. Two trial prompts said 'vertical 9:16' and still came out 16:9 because the setting was 16:9.

The shape of a good prompt

MiniMax H3
Write shots with hard-cut timestamps, then overall_soundscape (tag dialogue S1) and non_diegetic_music. Best at 8-12 s.
LTX-2.5
Write one flowing paragraph in the present tense with an explicit camera move and an Audio: part. Keep to one or two clear actions.

Cost, quality and time

Times are end to end on our GPU (an RTX 4060 Laptop GPU, 8 GB), model loading included, before any jobs ahead of yours. "Estimate" marks an option we have not measured yet.

OptionCreditsTypical timeSlowest seenWhat to expect
MiniMax H3 · Final · up to 12 sRecommended2221 min 31 s22 min 26 s2 runsBest picture and sound. Measured: 11.5 s clip 20 min 35 s, 12.25 s clip 22 min 26 s at 960x544. H3 is at its best at 8-12 s.
MiniMax H3 · Final · 13-15 s3231 min 28 s32 min 14 s2 runsSame quality, longer clip. Measured: two 15 s clips took 30 min 42 s and 32 min 14 s.
MiniMax H3 · Draft · up to 12 s43 min 21 s3 min 21 s8-step turbo version of Final: near-identical composition, a little less detail. The app's own guide says about 10-15 min per 10 s clip. Time to be measured; this is a conservative estimate. Measured 201 s on our GPU (30 Oct 2026).
LTX-2.5 · Balanced · up to 12 s54 min 17 s4 min 17 s1 runMuch faster and cheaper, sharper resolution (1280x704), less cinematic than H3 and weaker on fast aircraft. Measured once: 12 s clip in 4 min 17 s.
LTX-2.5 · Balanced · 13-20 s87 min 24 s7 min 24 s1 runThe longest clips we offer. Measured once: 20 s clip in 7 min 24 s.
LTX-2.5 · Preview · up to 20 s158 s58 sFirst stage only, at half size: a cheap way to check a prompt before paying for Balanced or H3. Time to be measured; this is a conservative estimate. Measured 58 s on our GPU (30 Oct 2026).

Combine with other tools

Step-by-step recipes that use Cinematic video with sound with other studio tools, with the credits each step costs.

The settings, explained

Promptrequired
MiniMax H3: shots with timestamps, then overall_soundscape and non_diegetic_music. LTX-2.5: one present-tense paragraph with a camera move and an 'Audio:' part.Accepts: 20-4000 characters.
Length in seconds
Accepts: set by the option you choose.
Frame shape
Accepts: 16:9 wide, 9:16 tall.
First frame (optional)
Optional picture the clip starts from (MiniMax H3 only). Use only pictures you own.Accepts: JPG, PNG, WEBP, up to 8 MB.Only with MiniMax H3.

Create

You answer for your prompt and your choice of model and settings; the safety filter still checks every job, and a refused prompt costs nothing.

Choose the model and quality
Describe what you want

We write it in the shape this model works best with, timed for 10 seconds. You can edit the result before you send it.

Camera (optional)

Mood (optional)

Light and setting (optional)

Sound (optional)

Music (optional)

Type your idea above first: the prompt is written from it, in your browser.

Prompt shape for this model: Write shots with hard-cut timestamps, then overall_soundscape (tag dialogue S1) and non_diegetic_music. Best at 8-12 s.

Start from a prompt that worked

5-12 seconds for this option.

Frame shape

First frame (optional)

Optional picture the clip starts from (MiniMax H3 only). Use only pictures you own.

22 creditsabout 22-23 min on our GPU, plus any jobs ahead

You'll have 7 days to download it; then it's deleted.

Results are kept 7 days. Credits come back only if the job fails or never starts.

The line right now

Reading the line…

Credits are held when you send a job and spent when the result is ready. They come back automatically if the job fails, is refused or has not started within 24 hours. Results are kept 7 days, or 48 hours after your first download, then deleted; a result that expires unused is not refunded. Pricing and refunds

Frequently asked questions

Which model should I pick?

MiniMax H3 Final for the best picture and sound (22 credits up to 12 s). LTX-2.5 Balanced when speed and price matter more (5 credits up to 12 s).

Does the clip really have sound?

Yes. Both models generate the audio track with the picture: effects, music and short spoken lines you write in quotes.

How long will I wait?

MiniMax H3 Final took 20-32 minutes per clip in our runs, LTX-2.5 Balanced 4-7.5 minutes, plus any jobs ahead of you. The page shows the live estimate before you pay.

Can I start from my own picture?

Yes, with MiniMax H3: upload a first frame. Use only pictures you own or have the right to use.

What if the clip fails?

Credits are only spent when the clip is delivered. A failed or refused job returns them automatically.

More in the Creative Studio